The DOI matches its Crossref record; 132 of the 132 references it cites were found in a registry or an index; 1 of the 78 sources its claims cite were checked for support.
Gradient-based learning applied to document recognition
Open accessAbstract
Multilayer neural networks trained with the back-propagation algorithm constitute the best example of a successful gradient based learning technique. Given an appropriate network architecture, gradient-based learning algorithms can be used to synthesize a complex decision surface that can classify high-dimensional patterns, such as handwritten characters, with minimal preprocessing. This paper reviews various methods applied to handwritten character recognition and compares them on a standard handwritten digit recognition task. Convolutional neural networks, which are specifically designed to deal with the variability of 2D shapes, are shown to outperform all other techniques. Real-life document recognition systems are composed of multiple modules including field extraction, segmentation recognition, and language modeling. A new learning paradigm, called graph transformer networks (GTN), allows such multimodule systems to be trained globally using gradient-based methods so as to minimize an overall performance measure. Two systems for online handwriting recognition are described. Experiments demonstrate the advantage of global training, and the flexibility of graph transformer networks. A graph transformer network for reading a bank cheque is also described. It uses convolutional neural network character recognizers combined with global training techniques to provide record accuracy on business and personal cheques. It is deployed commercially and reads several million cheques per day.
Overview
AI snapshot
Written from the paper’s full text.
TL;DR
The paper argues that gradient-based learning, especially convolutional neural networks and graph transformer networks, can replace much hand-crafted feature extraction and heuristic module tuning in document recognition. It reports strong handwritten digit results and demonstrates globally trained systems for online handwriting and commercial check reading.
Key findings
- On MNIST, LeNet-5 reached 0.95% test error on the regular database and 0.8% when trained with artificially distorted examples; boosted LeNet-4 achieved the best raw error at 0.7%.
- Convolutional networks outperformed several alternatives on the same digit benchmark, including linear classifiers (12%), pairwise linear classifiers (7.6%), K-NN (5.0%), PCA plus polynomial classifier (3.3%), RBF network (3.6%), one-hidden-layer multilayer networks (4.5–4.7%), two-hidden-layer networks (2.95–3.05%), LeNet-1 (1.7%), LeNet-4 (1.1%), and tangent distance (1.1%).
- Global word-level training improved online handwriting recognition: for HOS with a 25,461-word dictionary, word/character errors fell from 4.6%/2.0% to 3.2%/1.4%; for SDNN/HMM without dictionary constraints, they fell from 38%/12.4% to 26%/8.2%.
- A GTN-based check reader on 646 machine-printed business checks achieved 82% correct, 1% error, and 17% rejects, compared with 68% correct, 1% error, and 31% rejects for the previous system; it was deployed commercially and read millions of checks per day.
Methods
- The digit comparison used the MNIST database, built from NIST SD-1 and SD-3, with 60,000 training patterns and a 10,000-pattern test subset (5,000 from each source), size-normalized and centered; LeNet-5 is a seven-layer convolutional network with convolutional, subsampling, fully connected, and RBF output layers.
- The paper introduces graph transformer networks, where modules take and produce directed graphs whose arcs carry numerical information; training uses backpropagation through differentiable graph transformations, including Viterbi, forward, and discriminative forward criteria.
- Applications include online handwriting recognition with HOS and SDNN/HMM architectures, and a check amount reading cascade combining field location, segmentation, LeNet-5 character recognition, grammar composition, and Viterbi selection.
Limitations
- Stated: the paper notes that the absence of harmful local minima in multilayer networks remains theoretically mysterious; that more training data would likely improve LeNet-5 accuracy; that SDNN had not yet outperformed HOS; and that global training had only a minor effect in the check-reading task because only a small subset of parameters was trained globally.
- Evident from the text: the main quantitative benchmark is limited to isolated handwritten digits, so the reported comparisons may not transfer directly to all document recognition tasks.
- Evident from the text: the check-reading results come from a specific 646-check business-check test set and a commercial deployment, with limited detail for independent reproduction.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
244 Own interpretation · 84 No basis given · 66 Own result · 25 Not checked yet · 20 No full text of the sources · 8 Own method · 1 Sources not found · accuracy not yet measured
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Test error was 1.1%.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 8) LeNet-4: Experiments with
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The error rate on the regular test set was 7.6%.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Comparison with Other Classifiers
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However the memory requirement and recognition time are large: the complete 60 000 20 20 pixel training images (about 24 megabytes at one byte per pixel) must be available at run time.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Comparison with Other Classifiers
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“When deslanted images were used, the test error jumped down to 1.6% for a network with 300 hidden units.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 5) One-Hidden-Layer Fully Connected Multilayer NN:
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“On the regular test set the error rate was 5.0%.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Comparison with Other Classifiers
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The results clearly demonstrate the advantages of training a recognizer at the word level, rather than training it on presegmented, hand-labeled, isolated characters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > I. INTRODUCTION
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However, Appendix B shows that despite many claims to the contrary in the literature, the usefulness of these second-order methods to large learning machines is very limited.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Gradient-Based Learning
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“While this phenomenon is very common, it was not observed in our case as the learning curves in Fig. 5 show.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“SD-1 contains 58 527 digit images written by 500 different writers.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Database: The Modified NIST Set
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Convolutional neural networks, which are specifically designed to deal with the variability of two dimensional (2-D) shapes, are shown to outperform all other techniques.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Gradient-Based Learning Applied to Document Recognition
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- It may state two separate claims as one. 119
- Probably not a substantial claim. 154
- The quote only partly states this claim. 6
None of the sources cited for this claim have been checked yet.
In the paper
“On the other hand, if the probabilistic assumptions in an HMM (or other probabilistic model) are not realistic, discriminative training, discussed in Section VI, can improve performance as this has been clearly shown for speech recognition systems [48]- [50], [107], [108].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. GTN and HMM's
What the sources say
- Could not verifycites [48] 48. T. G. Dietterich and G. Bakiri, "Solving multiclass learning problems via error-correcting output codes," J. Artificial Intell. Res., vol. 2, pp. 263-286, 1995. sourceExperimental — accuracy not yet measured
The Introduction discusses phoneme recognition as a multiclass task, and Section 2.1 describes the NETtalk pronunciation dataset. Neither passage addresses unrealistic HMM assumptions or whether discriminative training improves speech-recognition performance; searches for those terms found no supporting passage in our copy. (LLM, conf=0.75)
- Not checked — no full text of the sourcecites [49] 49. L. R. Bahl, P. F. Brown, P. V. de Souza, and R. L. Mercer, "Maximum mutual information of hidden Markov model pa- rameters for speech recognition," in Proc. Int. Conf. Acoustics, Speech, Signal Processing, 1986, pp. 49-52. source
- Not checked yetcites [50] 50. "Speech recognition with continuous-parameter hidden Markov models," Comput., Speech Language, vol. 2, pp. 219-234, 1987.
- Not checked — no full text of the sourcecites [107] 107. M. Mohri, F. C. N. Pereira, and M. Riley, A Rational Design for a Weighted Finite-State Transducer Library (Lecture Notes in Computer Science). New York: Springer Verlag, 1997. source
- Not checked — no full text of the sourcecites [108] 108. M. Rahim, C. H. Lee, and B. H. Juang, "Discriminative utterance verification for connected digits recognition," IEEE Trans. Speech Audio Processing, vol. 5, pp. 266-277, 1997. source
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Could not verify
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“On the regular data, the error rate is 12%.The network has 7850 free parameters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Comparison with Other Classifiers
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Similarly, the new test set was completed with SD-3 examples starting at pattern #35 000 to make a full set with 60 000 test patterns.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Database: The Modified NIST Set
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one. The quote only partly states this claim.
- Rests on an assumption: Drawing sensible conclusions from learning experiments requires that the result be independent of the choice of training set and test among the complete set of samples.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Various combinations of sigmoid units, linear units, gradient descent learning, and learning by directly solving linear systems gave similar results.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Comparison with Other Classifiers
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The quote only partly states this claim.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Error on the regular test set was 4.7% for a network with 300 hidden units and 4.5% for a network with 1000 hidden units.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 5) One-Hidden-Layer Fully Connected Multilayer NN:
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Thus we had two sets with nearly 30 000 examples each.The new training set was completed with enough examples from SD-3, starting at pattern #0, to make a full set of 60 000 training patterns.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Database: The Modified NIST Set
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: Drawing sensible conclusions from learning experiments requires that the result be independent of the choice of training set and test among the complete set of samples.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“It is interesting to note that the network effectively sees each individual sample only twice over the course of these 20 passes.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“LeNet-5 comprises seven layers, not counting the input, all of which contain trainable parameters (weights).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. LeNet-5
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“LeNet-4 contains about 260 000 connections and has about 17 000 free parameters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 8) LeNet-4: Experiments with
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In the experiments described here, we only used a subset of 10 000 test images (5,000 from SD-1 and 5,000 from SD-3), but we used the full 60 000 training samples.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Database: The Modified NIST Set
plainn = 60,000- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“On the deslanted images, the test error rate is 8.4%.The network has 4010 free parameters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Comparison with Other Classifiers
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“It is deployed commercially and reads several million checks per day.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Gradient-Based Learning Applied to Document Recognition
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The test error rate of a 28 28-300-100-10 network was 3.05%, a much better result than the one-hidden-layer network, obtained using marginally more weights and connections.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 6) Two-Hidden-Layer Fully Connected Multilayer NN:
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Neither of those methods improved the raw error rate, although they did improve the rejection performance.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 8) LeNet-4: Experiments with
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Although about 100 000 multiply/add steps are required to evaluate LeNet-1, its convolutional nature keeps the number of free parameters to only about 2600.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 6) Two-Hidden-Layer Fully Connected Multilayer NN:
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The input is a 32 32 pixel image.This is significantly larger than the largest character in the database (at most 20 20 pixels centered in a 28 28 field).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. LeNet-5
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Experiments demonstrate the advantage of global training, and the flexibility of graph transformer networks.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Gradient-Based Learning Applied to Document Recognition
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Training with distorted patterns improved the performance somewhat: 2.50% error for the 28 28-300-100-10 network, and 2.45% for the 28 28-1000-150-10 network.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 6) Two-Hidden-Layer Fully Connected Multilayer NN:
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The error on the regular test set was 3.3%.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Comparison with Other Classifiers
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The quote only partly states this claim.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“In contrast to SD-3, where blocks of data from each writer appeared in sequence, the data in SD-1 is scrambled.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Database: The Modified NIST Set
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“It is reading millions of checks per month in several banks across the United States.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > I. INTRODUCTION
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“This system is in commercial use in the NCR Corporation line of check recognition systems for the banking industry.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > I. INTRODUCTION
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“However, SD-3 is much cleaner and easier to recognize than SD-1.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Database: The Modified NIST Set
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The test error rate was 0.7%, the best of any of our classifiers.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 8) LeNet-4: Experiments with
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one. The quote only partly states this claim.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The original black and white (bilevel) images were size normalized to fit in a 20 20 pixel box while preserving their aspect ratio.The resulting images contain grey levels as result of the antialiasing (image interpolation) technique used by the normalization algorithm.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Database: The Modified NIST Set
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“K-NN's and tangent distance classifier have essentially zero training time.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Most methods require only about 1 byte per variable for adequate performance.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The enhanced versions of LeNet-4 did better than the original LeNet-4, even though the raw accuracies were identical.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As can be seen on these examples, the network exhibits striking invariance and noise resistance properties.”
- Also stated in: § B. Experiments with SDNN · plain
“While some authors have argued that invariance requires more sophisticated models than feedforward NN's [87], LeNet-5 exhibits these properties to a large extent.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Experiments with SDNN
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Also stated in: § B. Experiments with SDNN · plain
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The neural-net methods run much faster and require much less space than memorybased techniques.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The gradients back propagated through the right half of the GTN are multiplied by 1, since contributes to the loss with a negative sign.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“If an arc appears in but not in the gradient is 1.The arc should have had a lower penalty to make it to”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“If an arc is in but not in the gradient is The arc had a low penalty, but it should have had a higher penalty since it is not part of the desired answer.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Figs. 25 and 26 show a few examples of successful recognitions of multiple characters by the LeNet-5 SDNN.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Experiments with SDNN
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In our work on building large-scale handwriting recognition systems, we have found that these systems could be developed and designed much more easily and quickly by viewing the system as a networks of modules that take one or several graphs as input and produce graphs as output.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. GTN's
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The RS-SVM requirements are within a factor of two of the convolutional networks, and the error rate is very close.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The gradients on arcs of get positive contributions from the left half and negative contributions from the right half.”
- Also stated in: § B. Discriminative Viterbi Training · plain
“The two contributions must be added since the penalties on arcs are sent to the two halves through a”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Also stated in: § B. Discriminative Viterbi Training · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“While the single-layer net, the pairwise net, and PCA quadratic net could be trained in less than an hour, the multilayer net training times were expectedly much longer, but only required 10-20 passes through the training set.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“(12) is always positive since the constrained graph is a subset of the paths in the interpretation graph, and the Viterbi algorithm selects the path with the lowest total penalty.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“In the ideal case, the two paths and coincide, and is zero.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“We find that boosting gives a substantial improvement in accuracy, with a relatively modest penalty in memory and computing expense.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Arcs in that appear neither in nor in have a gradient of zero.They do not contribute to the cost.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“the multilayer net training times were expectedly much longer, but only required 10-20 passes through the training set.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The success of this approach was demonstrated with a commercially deployed checkreading system that reads millions of business and personal checks per day: the generalized transduction engine resides in only a few hundred lines of code.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
boosteddirectSingle case- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“This can be compared to the performance of the previous system on the same test set: 68% correct, 1% errors, and 31% rejects.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“An independent test performed by systems integrators in 1995 showed the superiority of this system over other commercial courtesy amount reading systems.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“This algorithm is not believed to bring a tremendous increase in learning speed but it converges reliably without requiring extensive adjustments of the learning parameters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Additional information
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“It has been fielded in several banks across the United States since June 1996, and it has been reading millions of checks per day since then.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“When using the word normalization preprocessing instead of a character level normalization, error rates dropped to 4.6% and 2.0% for word and character errors respectively, i.e., a relative drop of 37% and 43% in word and character error respectively.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Experimental Results
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“With a 25 461-word dictionary, errors dropped from 4.6% and 2.0% word and character errors to 3.2% and 1.4%, respectively, after word-level training, i.e., a relative drop of 30.4% and 30.0%.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Experimental Results
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The additional cost of using this procedure over standard stochastic gradient descent is negligible.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Additional information
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The system that deals with handwritten or personal checks was based on earlier implementations that used the GTN concept in a restricted way.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Results
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“On 646 business checks that were automatically categorized as machine printed, the performance was 82% correctly recognized checks, 1% errors, and 17% rejects.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The system was integrated in NCR's line of check reading systems.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For the HOS system and a slightly improved architecture, without any dictionary constraints, the error rates dropped from 22.5% and 8.5% word and character error to 17% and 6.3% respectively, i.e., a relative drop of 24.4% and 25.6%.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Experimental Results
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The GTN methodology was used to build a check amount reading system that handles both personal checks and business checks.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Experimental Results
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
- Rests on an assumption: Checks contain at least two versions of the amount: the courtesy amount written with numerals and the legal amount written with letters.
- Rests on an assumption: On business checks, amounts are generally machine-printed, relatively easy to read but quite difficult to find due to the lack of standard for business check layout.
- Rests on an assumption: On personal checks, amounts are easy to find but much harder to read.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The Viterbi transformer finally selects the path with the lowest accumulated penalty corresponding to the best grammatically correct interpretations.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 5) The Viterbi Transformer:
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The recognizer classifies segment images into one of 95 classes (fully printable ASCII set) plus a rubbish class for unknown symbols or badly formed characters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. A GTN for Check Amount Recognition
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For the SDNN/HMM system, without any dictionary constraints, the error rates dropped from 38% and 12.4% word and character error to 26% and 8.2% respectively after word-level training, i.e., a relative drop of 32% and 34%.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Experimental Results
plaindirectn = 3,500- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Larger architectures did not necessarily perform better and required considerably more time to be trained.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Network Architecture
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The importance of global training was only minor in this task because the global training only concerned a small subset of the parameters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Results
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The basic idea of back propagation is that gradients can be computed efficiently by propagation from the output to the input.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Gradient Back Propagation
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The simplest output loss function that can be used with the above network is the maximum likelihood estimation criterion, which in our case is equivalent to the minimum mean squared error (MSE).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Loss Function
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“If the loss function measuring the performance can be made differentiable with respect to the system's tunable parameters we can find a local minimum of using gradient-based learning.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Globally Trainable Systems
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As stated earlier, all the units in a feature map share the same set of 25 weights and the same bias, so they detect the same feature at all possible locations on the input.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Convolutional Networks
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The rationale is that characters that are similar, and therefore confusable, such as uppercase “O,” lowercase “o,” and zero, lowercase “l” digit one, and square brackets and uppercase “I,” will have similar output codes.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. LeNet-5
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one. The quote only partly states this claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This operation is equivalent to a convolution, followed by an additive bias and squashing function, hence the name convolutional network.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Convolutional Networks
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“Gradient-based learning procedures have been used since the late 1950's, but they were mostly limited to linear systems [1].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Gradient Back Propagation
What the sources say
- Not checked — no full text of the sourcecites [1] 1. R. O. Duda and P. E. Hart, Pattern Classification and Scene Analysis. New York: Wiley, 1973.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Convolutional networks combine three architectural ideas to ensure some degree of shift, scale, and distortion invariance: 1) local receptive fields; 2) shared weights (or weight replication); and 3) spatial or temporal subsampling.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Convolutional Networks
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“The weight sharing technique has the interesting side effect of reducing the number of free parameters, thereby reducing the “capacity” of the machine and reducing the gap between test error and training error [34].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Convolutional Networks
What the sources say
- Not checked yetcites [34] 34. M. C. Mozer, The Perception of Multiple Objects: A Con- nectionist Approach. Cambridge, MA: MIT-Bradford Books, 1991.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Traditional multilayer NN's are a special case of the above where the state information is represented with fixed-sized vectors, and where the modules are alternated layers of matrix multiplications (the weights) and component-wise sigmoid functions (the neurons).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Globally Trainable Systems
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It is interesting to note that in many instances there is no need to explicitly compute the Jacobian matrix.The above formula uses the product of the Jacobian with a vector of partial derivatives, and it is often easier to compute this product directly without computing the Jacobian beforehand.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Globally Trainable Systems
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“The Lagrange formalism used in the control theory literature provides perhaps the best rigorous method for deriving back propagation [20] and for deriving generalizations of back propagation to recurrent networks [21] and networks of heterogeneous modules [22].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Gradient Back Propagation
What the sources say
- Not checked yetcites [20] 20. D. B. Parker, "Learning-logic," Sloan School Manage., MIT, Cambridge, MA, Tech. Rep., TR-47, Apr. 1985.
- Not checked — no full text of the sourcecites [21] 21. Y. LeCun, Modéles Connexionnistes de l'Apprentissage (Con- nectionist Learning Models), Ph.D. dissertation, UniversitéP . et M. Curie (Paris 6), June 1987.
- Not checked yetcites [22] 22. "A theoretical framework for back-propagation," in Proc. 1988 Connectionist Models Summer School, D. Touretzky, G. Hinton, and T. Sejnowski, Eds. Pittsburgh, PA: CMU, Morgan Kaufmann, 1988, pp. 21-28.
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Efficient learning algorithms can be devised when the gradient vector can be computed analytically (as opposed to numerically through perturbations).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Gradient-Based Learning
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It is worth pointing out that the components of those vectors are 1o r 1, which is well within the range of the sigmoid of F6, and therefore prevents those sigmoids from getting saturated.In fact, 1 and 1 are the points of maximum curvature of the sigmoids.This forces the F6 units to operate in their maximally nonlinear range.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. LeNet-5
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: Saturation of the sigmoids leads to slow convergence and ill-conditioning of the loss function.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The ability of multilayer networks trained with gradient descent to learn complex, high-dimensional, nonlinear mappings from large collections of examples makes them obvious candidates for image recognition tasks.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > II. CONVOLUTIONAL NEURAL NETWORKS FOR ISOLATED CHARACTER RECOGNITION
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Recognizing variable-length objects such as handwritten words using multimodule systems is best done if the modules manipulate directed graphs.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > I. INTRODUCTION
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It may seem paradoxical that gradients can be computed when the state information is represented by essentially discrete objects such as graphs, but that difficulty can be circumvented, as shown later.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Globally Trainable Systems
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“A formal version of this is called structural risk minimization [6], [7], and it is based on defining a sequence of learning machines of increasing capacity, corresponding to a sequence of subsets of the parameter space such that each subset is a superset of the previous subset.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Learning from Data
What the sources say
- Not checked — no full text of the sourcecites [6] 6. V. N. Vapnik, The Nature of Statistical Learning Theory. New York: Springer, 1995. source
- Not checked yetcites [7] 7. Statistical Learning Theory. New York: Wiley, 1998.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Gradient-based learning draws on the fact that it is generally much easier to minimize a reasonably smooth, continuous function than a discrete (combinatorial) function.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Gradient-Based Learning
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Therefore, when increasing the capacity there is a tradeoff between the decrease of and the increase of the gap, with an optimal value of the capacity that achieves the lowest generalization error”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Learning from Data
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It can be argued that the recent progress in the accuracy of speech and handwriting recognition systems can be attributed in large part to an increased reliance on learning techniques and large training data sets.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > I. INTRODUCTION
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“First, a noncomplete connection scheme keeps the number of connections within reasonable bounds.More importantly, it forces a break of symmetry in the network.Different feature maps are forced to extract different (hopefully complementary) features because they get different sets of inputs.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. LeNet-5
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“With local receptive fields neurons can extract elementary visual features such as oriented edges, endpoints, corners (or similar features in other signals such as speech spectrograms).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Convolutional Networks
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In such a model, the accuracy of the system depends upon the quality of the cuts generated by the heuristics, and on the ability of the recognizer to distinguish correctly segmented characters from pieces of characters, multiple characters, or otherwise incorrectly segmented characters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Learning in Real Handwriting Recognition Systems
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This last step is extremely tedious, time consuming, and almost certainly suboptimal.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Globally Trainable Systems
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
In the paper
“Convolutional NN's [2], introduced in Section II, are an example of specialized NN architectures which incorporate knowledge about the invariances of twodimensional (2-D) shapes by using local connection patterns and by imposing constraints on the weights.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > I. INTRODUCTION
What the sources say
- Not checked — no full text of the sourcecites [2] 2. Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, and L. D. Jackel, "Backpropagation applied to handwritten zip code recognition," Neural Computation, vol. 1, no. 4, pp. 541-551, Winter 1989. source
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“A fully connected first layer with, e.g., one hundred hidden units in the first layer would already contain several tens of thousands of weights.Such a large number of parameters increases the capacity of the system and therefore requires a larger training set.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > II. CONVOLUTIONAL NEURAL NETWORKS FOR ISOLATED CHARACTER RECOGNITION
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Fig. 8 shows all 82 misclassified test examples.some of those examples are genuinely ambiguous, but several are perfectly identifiable by humans, although they are written in an under-represented style.This shows that further improvements are to be expected with more training data.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Results
plainconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Minimizing in effect limits the capacity of the accessible subset of the parameter space, thereby controlling the tradeoff between minimizing the training error and minimizing the expected gap between the training error and test error.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Learning from Data
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“While more automatic learning is beneficial, no learning technique can succeed without a minimal amount of prior knowledge about the task.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > I. INTRODUCTION
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“An interesting property of convolutional layers is that if the input image is shifted, the feature map output will be shifted by the same amount, but it will be left unchanged otherwise.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Convolutional Networks
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“Historically, the need for appropriate feature extractors was due to the fact that the learning techniques used by the classifiers were limited to low-dimensional spaces with easily separable classes [1].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > I. INTRODUCTION
What the sources say
- Not checked — no full text of the sourcecites [1] 1. R. O. Duda and P. E. Hart, Pattern Classification and Scene Analysis. New York: Wiley, 1973.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Once a feature has been detected, its exact location becomes less important.Only its approximate position relative to other features is relevant.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Convolutional Networks
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The other feature maps in the layer use different sets of weights and biases, thereby extracting different types of local features.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Convolutional Networks
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“The values of the input pixels are normalized so that the background level (white) corresponds to a value of and the foreground (black) corresponds to 1.175.This makes the mean input roughly zero and the variance roughly one, which accelerates learning [46].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. LeNet-5
What the sources say
- Not checked — no full text of the sourcecites [46] 46. J. Bromley, J. W. Bentz, L. bottou, I. Guyon, Y. LeCun, C. Moore, E. Säckinger, and R. Shah, "Signature verification using a siamese time delay neural network," Int. J. Pattern Recognit. Artificial Intell., vol. 7, no. 4, pp. 669-687, Aug. 1993. source
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In practice, the performance of the system on a training set is of little interest.The more relevant measure is the error rate of the system in the field, where it would be used in practice.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Learning from Data
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“A completely rigorous derivation in more general cases can be done using Lagrange functions [20]- [22].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Globally Trainable Systems
What the sources say
- Not checked yetcites [20] 20. D. B. Parker, "Learning-logic," Sloan School Manage., MIT, Cambridge, MA, Tech. Rep., TR-47, Apr. 1985.
- Not checked — no full text of the sourcecites [21] 21. Y. LeCun, Modéles Connexionnistes de l'Apprentissage (Con- nectionist Learning Models), Ph.D. dissertation, UniversitéP . et M. Curie (Paris 6), June 1987.
- Not checked yetcites [22] 22. "A theoretical framework for back-propagation," in Proc. 1988 Connectionist Models Summer School, D. Touretzky, G. Hinton, and T. Sejnowski, Eds. Pittsburgh, PA: CMU, Morgan Kaufmann, 1988, pp. 21-28.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“One of the most difficult problems in handwriting recognition, however, is not only to recognize individual characters, but also to separate out characters from their neighbors within the word or sentence, a process known as segmentation.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Learning in Real Handwriting Recognition Systems
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In most cases, the information carried from module to module is best represented as graphs with numerical information attached to the arcs.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Globally Trainable Systems
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However, as stated earlier, the state information in complex recognition system is best represented by graphs with numerical information attached to the arcs.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Globally Trainable Systems
plain- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“A better alternative would be to somehow train the entire system so as to minimize a global error measure such as the probability of character misclassifications at the document level.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Globally Trainable Systems
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In principle, a fully connected network of sufficient size could learn to produce outputs that are invariant with respect to such variations.However, learning such a task would probably result in multiple units with similar weight patterns positioned at various locations in the input so as to detect distinctive features wherever they appear on the input.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > II. CONVOLUTIONAL NEURAL NETWORKS FOR ISOLATED CHARACTER RECOGNITION
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In the case of multilayer NN's, a good way to incorporate knowledge is to tailor its architecture to the task.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > I. INTRODUCTION
plainconditional- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In addition, elementary feature detectors that are useful on one part of the image are likely to be useful across the entire image.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Convolutional Networks
hedgedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In convolutional networks, as described below, shift invariance is automatically obtained by forcing the replication of weight configurations across space.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > II. CONVOLUTIONAL NEURAL NETWORKS FOR ISOLATED CHARACTER RECOGNITION
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Another reason for using such distributed codes, rather than the more common “1 of N” code (also called place code or grandmother cell code) for the outputs is that nondistributed codes tend to behave badly when the number of classes is larger than a few dozen.The reason is that output units in a nondistributed code must be off most of the time.This is quite difficult to achieve with sigmoid units”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. LeNet-5
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Unfortunately, no such preprocessing can be perfect: handwriting is often normalized at the word level, which can cause size, slant, and position variations for individual characters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > II. CONVOLUTIONAL NEURAL NETWORKS FOR ISOLATED CHARACTER RECOGNITION
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Learning these weight configurations requires a very large number of training instances to cover the space of possible variations.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > II. CONVOLUTIONAL NEURAL NETWORKS FOR ISOLATED CHARACTER RECOGNITION
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In the ideal case, the penalties of incorrect paths are infinitely large, therefore the two penalties coincide and is zero.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discriminative Forward Training
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“More generally, the same idea can be used in all situations where a learning machine must choose between discrete alternative interpretations.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discriminative Forward Training
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“When plenty of data are available, many methods can attain respectable accuracy.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
plainconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Third, a word or character string cannot be perfectly size-normalized.Individual characters within a string may have widely varying sizes and baseline positions.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > VII. MULTIPLE OBJECT RECOGNITION:S PACE DISPLACEMENT NEURAL NETWORK
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Translated in terms of penalties, the penalty of an interpretation should be the negative logarithm of the sum of the negative exponentials of the penalties of the individual paths.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Forward Scoring and Forward Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
- The authors state this as an assumption: Penalties can be interpreted as negative log posteriors, so that posterior probabilities for an interpretation combine by summing and penalties combine as the negative logarithm of the sum of negative exponentials.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The sum of those posteriors for all the possible label sequences is one.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Remarks on Discriminative Training
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- The authors state this as an assumption: A particular label sequence is given as the desired sequence to the GTN.
In the paper
“Popular HMM learning methods, such as the Baum-Welsh algorithm, rely on this normalization.The normalization cannot be preserved when nongenerative models such as NN's are integrated into the system.Other techniques, such as discriminative training methods, must be used in this case.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > VI. GLOBAL TRAINING FOR GRAPH TRANSFORMER NETWORKS
What the sources say
- Not checked — no full text of the sourcecites [28] 28. O. Matan, C. J. C. Burges, Y. LeCun, and J. S. Denker, "Multi- digit recognition using a space displacement neural network," vol. 4, in Neural Information Processing Systems, J. M. Moody, S. J. Hanson, and R. P. Lippman, Eds. San Mateo, CA: Morgan Kaufmann, 1992.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In this framework, there is no qualitative difference between trainable parameters (W1, W2 in the figure), external inputs and outputs (Z,D,E), and intermediate state variables (X1, X2, X3, X4, X5).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > IV. MULTIMODULE SYSTEMS AND GRAPH TRANSFORMER NETWORKS
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
None of the sources cited for this claim have been checked yet.
In the paper
“The best way to justify it theoretically is through the use of Lagrange functions [21], [22].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. An Object-Oriented Approach
What the sources say
- Not checked — no full text of the sourcecites [21] 21. Y. LeCun, Modéles Connexionnistes de l'Apprentissage (Con- nectionist Learning Models), Ph.D. dissertation, UniversitéP . et M. Curie (Paris 6), June 1987.
- Not checked yetcites [22] 22. "A theoretical framework for back-propagation," in Proc. 1988 Connectionist Models Summer School, D. Touretzky, G. Hinton, and T. Sejnowski, Eds. Pittsburgh, PA: CMU, Morgan Kaufmann, 1988, pp. 21-28.
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The collapse does not occur in HMM-based speech recognition systems because they are generative systems that produce normalized likelihoods for the input data (more on this later).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Segmentation algorithms are rarely perfect and often leave extraneous marks in character images (noise, underlines, neighboring characters), or sometimes cut characters too much and produce incomplete characters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Invariance and Noise Resistance
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The only label information that is required by the above system is the sequence of desired character labels.No knowledge of the correct segmentation is required on the part of the supervisor, since it chooses among the segmentations in the interpretation graph the one that yields the lowest penalty.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- The authors state this as an assumption: The cut generation heuristic will include the correct set of cuts.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Finding the support vectors and the coefficients amounts to solving a high-dimensional quadratic minimization problem with linear inequality constraints.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 11) SVM:
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“All the penalties of the input graph have an influence on the forward penalty, but penalties that belong to low-penalty paths have a stronger influence.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Forward Scoring and Forward Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A claim is checked only against the cited paper’s full text, never its abstract, and none is available for its sources.
In the paper
“The idea has been applied to face location [93], address block location on envelopes [94], and hand tracking in video [95].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Object Detection and Spotting with SDNN
What the sources say
- Not checked — no full text of the sourcecites [93] 93. Y. Bengio and Y. LeCun, "Word normalization for on-line handwritten word recognition," in Proc. IEEE Int. Conf. Pattern Recognition, Jerusalem, 1994.
- Not checked — no full text of the sourcecites [94] 94. R. Vaillant, C. Monrocq, and Y. LeCun, "Original approach for the localization of objects in images," Proc. Inst. Elect. Eng., vol. 141, no. 4, pp. 245-250, Aug. 1994. source
- Not checked — no full text of the sourcecites [95] 95. R. Wolf and J. Platt, "Postal address block location using a convolutional locator network," in Advances in Neural Infor- mation Processing Systems 6, J. D. Cowan, G. Tesauro, and J. Alspector, Eds. San Mateo, CA: Morgan Kaufmann, 1994, pp. 745-752.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“More generally, fixed-size vectors lack flexibility for tasks in which the state must encode probability distributions over sequences of vectors or symbols, as is the case in linguistic processing.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. GTN's
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Recognizer outputs that have a nonzero gradient are part of the correct answer and will therefore have their value pushed down.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Unfortunately, those methods cannot be applied to discriminative training criteria, and one is reduced to using gradient-based methods.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Remarks on Discriminative Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In 1989 a recognizer as complex as LeNet-5 would have required several weeks' training and more data than were available and was therefore not even considered.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In fact, assuming that a particular label sequence is given as the”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Remarks on Discriminative Training
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- The authors state this as an assumption: A particular label sequence is given as the desired sequence to the GTN.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This is somewhat equivalent to modeling the output of an SDNN with an HMM.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Global Training of SDNN
hedgedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Another way to avoid the collapse is to train the whole system with respect to a discriminative training criterion, such as maximizing the conditional probability of the correct interpretations (correct sequence of class labels) given the input image.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
None of the sources cited for this claim have been checked yet.
In the paper
“SDNN's have been used in combination with HMM's or other elastic matching methods for handwritten word recognition [91], [92].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Global Training of SDNN
What the sources say
- Not checked yetcites [91] 91. C. Dugust, L. Devillers, and X. Aubert, "Combining TDNN and HMM in a hybrid system for improved continuous-speech recognition," IEEE Trans. Speech Audio Processing, vol. 2, pp. 217-224, Jan. 1994.
- Not checked — no full text of the sourcecites [92] 92. O. Matan, H. S. Baird, J. Bromley, C. J. C. Burges, J. S. Denker, L. D. Jackel, Y. LeCun, E. P. D. Pednault, W. Satterfield, C. E. Stenard, and T. J. Thompson, "Reading handwritten digits: A ZIP code recognition system," IEEE Trans. Comput., vol. 25, no. 7, pp. 59-63, July 1992. source
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The function of this module is differentiable everywhere, except on the switching surface which is a set of measure zero.Interestingly, this function is continuous and reasonably regular, and that is sufficient to ensure the convergence of a gradient-based learning algorithm.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Special Modules
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As computer technology improves, larger capacity recognizers become feasible.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As in the previous example, a discriminative criterion must be used, because using a nondiscriminative criterion could result in a collapse effect if the network's output RBF are adaptive.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Global Training of SDNN
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Discriminative training can also be seen as another example of error correction procedure, which tends to minimize the difference between the desired output computed in the left half of the GTN in Fig. 20 and the actual output computed in the right half of Fig. 20.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“but if one path”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Forward Scoring and Forward Training
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Therefore, many systems have resorted to normalizing the images at the level of fields or words.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Invariance and Noise Resistance
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The training therefore concentrates on examples of images which yield a classification error, and furthermore, it concentrates on the pieces of the image which cause that error.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discriminative Forward Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“A modification of the training criterion can circumvent the collapse problem described above and at the same time produce more reliable confidence values.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“While this module is not differentiable with respect to the switching input, it is differentiable with respect to the regular inputs.Therefore the overall function of a system that includes such modules will be differentiable with respect to its parameters as long as the switching input does not depend upon the parameters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Special Modules
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However, it is almost impossible to devise image analysis techniques that will infallibly segment naturally written sequences of characters into well formed characters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > V. MULTIPLE OBJECT RECOGNITION: HOS
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The answer is that large and complex trainable systems need to be built out of simple, specialized modules.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > IV. MULTIMODULE SYSTEMS AND GRAPH TRANSFORMER NETWORKS
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Because of the convolutional nature of the network, units in the two instances that look at identical locations on the input have identical outputs, therefore their states do not need to be computed twice.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > VII. MULTIPLE OBJECT RECOGNITION:S PACE DISPLACEMENT NEURAL NETWORK
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“All the recognizer instances share the same parameter vector, since they are merely clones of each other”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“If, on the other hand, the recognizer's output layer contains RBF units with fixed parameters, then there is no such trivial solution.”
- Also stated in: § A. Viterbi Training · plain
“In this case, the complete collapse described above does not occur.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Also stated in: § A. Viterbi Training · plain
In the paper
“Although separate training is simple, it requires additional supervision information that is often lacking or incomplete (the correct segmentation and the labels of incorrect candidate segments).Furthermore, it can be shown that separate training is suboptimal [67].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > VI. GLOBAL TRAINING FOR GRAPH TRANSFORMER NETWORKS
What the sources say
- Not checked yetcites [67] 67. Y. LeCun, L. Bottou, and Y. Bengio, "Reading checks with graph transformer networks," in Proc. IEEE Int. Conf. Acous- tics, Speech, Signal Processing. Munich, Germany, vol. 1, 1997, pp. 151-154,.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“For example, modeling the conditional distribution of the classes given the input image is more discriminative (focusing more on the classification surface) than having a separate generative model of the input data associated to each class (which, with class priors, yields the whole joint distribution of classes and inputs).”
- Also stated in: § B. Discriminative Viterbi Training · plain
“This is because the conditional approach does not need to assume a particular form for the distribution of the input data.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Also stated in: § B. Discriminative Viterbi Training · plain
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“to use the discriminative forward loss function because it causes less numerical problems during the optimization.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Remarks on Discriminative Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
In the paper
“Single chip mixed analog-digital implementations of LeNet-5's predecessors have been shown to operate at speeds in excess of 1000 characters per second [64].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
What the sources say
- Not checked — no full text of the sourcecites [64] 64. C. J. C. Burges and B. Schoelkopf, "Improving the accuracy and speed of support vector machines," in Advances in Neural Information Processing Systems 9, M. Jordan, M. Mozer, and T. Petsche, Eds. Cambridge, MA: MIT, 1997.
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Convolutional networks are particularly well suited for recognizing or rejecting shapes with widely varying size, position, and orientation, such as the ones typically produced by heuristic segmenters in real-world string recognition systems.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Invariance and Noise Resistance
boostedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In principle, derivatives can be backpropagated through any arrangement of functional modules, as long as we can compute the product of the Jacobians of those modules by any vector.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > IV. MULTIMODULE SYSTEMS AND GRAPH TRANSFORMER NETWORKS
plainconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The GTN techniques presented below simplify and generalize the global training methods developed for speech recognition.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > VI. GLOBAL TRAINING FOR GRAPH TRANSFORMER NETWORKS
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The discriminative Viterbi algorithm does not have the flaws of the nondiscriminative version, but there are problems nonetheless.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“Globally trained, variablesize TDNN/HMM hybrids have been used for speech recognition and online handwriting recognition [67], [77], [89], [90].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Global Training of SDNN
What the sources say
- Not checked yetcites [67] 67. Y. LeCun, L. Bottou, and Y. Bengio, "Reading checks with graph transformer networks," in Proc. IEEE Int. Conf. Acous- tics, Speech, Signal Processing. Munich, Germany, vol. 1, 1997, pp. 151-154,.
- Not checked yetcites [77] 77. X. Driancourt and L. Bottou, "MLP, LVQ and DP: Comparison & cooperation," in Proc. Int. Joint Conf. Neural Networks, Seattle, WA, vol. 2, 1991, pp. 815-819.
- Not checked — no full text of the sourcecites [89] 89. B. Boser, E. Säckinger, J. Bromley, Y. LeCun, and L. Jackel, "An analog neural network processor with programmable topol- ogy," IEEE J. Solid-State Circuits, vol. 26, pp. 2017-2025, Dec. 1991. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The same formalism can be used to extend the procedures to networks with recurrent connections.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. An Object-Oriented Approach
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Another problem with Viterbi training is that the penalty of the answer cannot be used reliably as a measure of confidence because it does not take low-penalty (or highscoring) competing answers into account.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
In the paper
“Given an interpretation, there is a well-known method, called the forward algorithm for computing the above quantity efficiently [28].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Forward Scoring and Forward Training
What the sources say
- Not checked — no full text of the sourcecites [28] 28. O. Matan, C. J. C. Burges, Y. LeCun, and J. S. Denker, "Multi- digit recognition using a space displacement neural network," vol. 4, in Neural Information Processing Systems, J. M. Moody, S. J. Hanson, and R. P. Lippman, Eds. San Mateo, CA: Morgan Kaufmann, 1992.
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Although an explicit”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Remarks on Discriminative Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“Experiments in speech recognition with hybrids of NN's and HMM's also showed marked improvements brought by global training [29], [67], [77], [84].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Remarks on Discriminative Training
What the sources say
- Not checked — no full text of the sourcecites [29] 29. L. R. Rabiner, "A tutorial on hidden Markov models and selected applications in speech recognition," Proc. IEEE, vol. 77, pp. 257-286, Feb. 1989. source
- Not checked yetcites [67] 67. Y. LeCun, L. Bottou, and Y. Bengio, "Reading checks with graph transformer networks," in Proc. IEEE Int. Conf. Acous- tics, Speech, Signal Processing. Munich, Germany, vol. 1, 1997, pp. 151-154,.
- Not checked yetcites [77] 77. X. Driancourt and L. Bottou, "MLP, LVQ and DP: Comparison & cooperation," in Proc. Int. Joint Conf. Neural Networks, Seattle, WA, vol. 2, 1991, pp. 815-819.
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The above training procedure can be equivalently formulated in term of HMM.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Global Training of SDNN
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Note that the forward penalty of the constrained graph is always larger or equal to the forward penalty of the unconstrained interpretation graph.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discriminative Forward Training
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“If the parameters of the RBF's are allowed to adapt, then the collapse problems reappear because the RBF centers can all converge to a single vector, and the underlying NN can learn to produce that vector and ignore the input.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This flat spot is a saddle point, but it is attractive in almost all directions and is very difficult to get out of using gradient-based minimization procedures.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Viterbi Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Discriminative training procedures can be seen as attempting to build appropriate separating surfaces between classes rather than to model individual classes independently of each other.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Discriminative Viterbi Training
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“Boser et al. [62] realized that any polynomial of degree in this”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > 11) SVM:
What the sources say
- Not checked — no full text of the sourcecites [62] 62. P. Simard, Y. LeCun, and J. Denker, "Efficient pattern recog- nition using a new transformation distance," in Advances in Neural Information Processing Systems, vol. 5, S. Hanson, J. Cowan, and L. Giles, Eds. San Mateo, CA: Morgan Kauf- mann, 1993.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Although combining penalties of different nature seems highly heuristic, the GTN training procedure will tune the penalties and take advantage of this combination anyway.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Recognition Transformer and Viterbi Transformer
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Also, distortion models can be used to increase the effective size of a data set without actually requiring to collect more data.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discussion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The forward penalty is always lower than the cumulated penalty on any of the paths”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Forward Scoring and Forward Training
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“NN's and many other standard pattern recognition techniques can be formulated in terms of multimodular systems trained with gradient-based learning.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Special Modules
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However the rapid progress of conventional processor technology with reduced-precision vector arithmetic instructions (such as Intel's MMX) make the success of specialized hardware hypothetical at best.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Experiments with SDNN
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
In the paper
“The Viterbi transformer owes its name to the famous Viterbi algorithm [70], an application of the principle of dynamic programming to find the shortest path in a graph efficiently.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Recognition Transformer and Viterbi Transformer
What the sources say
- Not matchedcites [70] 70. T. M. Breuel, "A system for the off-line recognition of hand- written text," in Proc. IEEE ICPR'94, Jerusalem, pp. 129-134.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Not matched
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We prefer to use the discriminative forward loss function because it causes less numerical problems during the optimization.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Remarks on Discriminative Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Enforcing the normalization of the probabilistic quantities while performing gradientbased learning is complex, inefficient, time consuming, and creates ill-conditioning of the loss-function.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Remarks on Discriminative Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“While this guarantees that stray marks will not be blown up into characterlooking images, this also creates wide variations of the size and vertical position of characters after segmentation.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Invariance and Noise Resistance
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“While the penalty of the Viterbi path is perfectly appropriate for the purpose of recognition it gives only a partial picture of the situation.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Forward Scoring and Forward Training
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Good candidate locations for cuts can be found by locating minima in the vertical projection profile, or minima of the distance between the upper and lower contours of the word.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Segmentation Graph
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As training data becomes plentiful, as computers get faster, and as our understanding of learning algorithms improves, recognition systems will rely more and more of learning and their performance will improve.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it. It may state two separate claims as one.
In the paper
“The idea of transforming graphs into other graphs has received considerable attention in computer science through the concept of weighted finite-state transducers [86].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Previous Work
What the sources say
- Not checked — no full text of the sourcecites [86] 86. U. Bodenhausen, S. Manke, and A. Waibel, "Connectionist architectural learning for high performance character and speech recognition," in Proc. Int. Conf. Acoustics, Speech, Signal Pro- cessing, Minneapolis, MN, vol. 1, 1993, pp. 625-628. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The promising SDNN approach draws on the robustness and efficiency of convolutional NN's to avoid explicit segmentation altogether.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This paper presents a small number of examples of GT modules, but it is clear that the concept can be applied to many situations where the domain knowledge or the state information can be represented by graphs.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Note that the input resolution is nonetheless much less than for OCR because the angle and curvature provide more information than would a single grey level at each pixel.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Network Architecture
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Gradient-based learning applied to global discriminative loss functions guarantees optimal classification and rejection without the use of “hard to justify” principles that put strong constraints on the system architecture, often at the expense of performances.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The rationale behind this is that the overall gain of the squashing transformation is around one in normal operating conditions, and the interpretation of the state of the network is simplified.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > APPENDIX A PRECONDITIONS FOR FASTER CONVERGENCE
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Hand-truthing images to obtain segmented characters for training a character recognizer is expensive and does not take into account the way in which a whole document or sequence of characters will be recognized”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“Numerous authors in speech recognition have used gradient-based learning methods that integrate graphbased statistical models (notably HMM's) with acoustic recognition modules, mainly Gaussian mixture models, but also NN's [67], [78], [98], [99].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Previous Work
What the sources say
- Not checked yetcites [67] 67. Y. LeCun, L. Bottou, and Y. Bengio, "Reading checks with graph transformer networks," in Proc. IEEE Int. Conf. Acous- tics, Speech, Signal Processing. Munich, Germany, vol. 1, 1997, pp. 151-154,.
- Not checked — no full text of the sourcecites [78] 78. Y. Bengio, R. De Mori, G. Flammia, and R. Kompe, "Global optimization of a neural network-hidden Markov model hybrid," IEEE Trans. Neural Networks, vol. 3, pp. 252-259, March 1992. source
- Not matchedcites [98] 98. E. Osuna, R. Freund, and F. Girosi, "Training support vector machines: An application to face detection," in Proc. IEEE CVPR'96, pp. 130-136.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“GTN's have been shown to reduce the need for hand-crafted heuristics, manual labeling, and manual parameter tuning in document recognition systems.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“To enhance the robustness of the recognizer to variations in position, size, orientation, and other distortions, additional training data was generated by applying local affine transformations to the original characters.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Experimental Results
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The quote only partly states this claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Consequently, the second derivative of the error with respect to this parameter may be very large, while it can be quite small for other parameters elsewhere in the network.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Additional information
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“On the other hand, GTN's extend HMM's by allowing to combine in a well-principled framework multiple levels of processing, or multiple models (e.g., Pereira et al. have been using the transducer framework for stacking HMM's representing different levels of processing in automatic speech recognition [86]).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. GTN and HMM's
What the sources say
- Not checked — no full text of the sourcecites [86] 86. U. Bodenhausen, S. Manke, and A. Waibel, "Connectionist architectural learning for high performance character and speech recognition," in Proc. Int. Conf. Acoustics, Speech, Signal Pro- cessing, Minneapolis, MN, vol. 1, 1993, pp. 625-628. source
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Once the number of subsampling layers and the sizes of the kernels are chosen, the sizes of all the layers, including the input, are determined unambiguously.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Network Architecture
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Another example of transduction was mentioned in Section V: the path selector used in the HOS training GTN is implementable by a composition.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Standard Transduction
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Instead we train multimodule systems to optimize a global measure of performance, which does not require time consuming detailed handtruthing and yields significantly better recognition performance because it allows to train these modules to cooperate toward a common goal.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This is equivalent to minimizing the MSE between the previous layer and the center of the correct-class RBF.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Network Training
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“The network was also initially trained to reject noncharacters that resulted from segmentation errors.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Limited-storage Broyden-Fletcher-Goldfarb-Shanno's and conjugate gradients require only operations per update so they would appear appropriate.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Additional information
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
In the paper
“This confirms similar results obtained earlier in speech recognition [77].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Experimental Results
What the sources say
- Not checked yetcites [77] 77. X. Driancourt and L. Bottou, "MLP, LVQ and DP: Comparison & cooperation," in Proc. Int. Joint Conf. Neural Networks, Seattle, WA, vol. 2, 1991, pp. 815-819.
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In the kind of systems we have discussed until now (and the application described in Section X), much of the knowledge about the structure of the graph that is produced by a GT is determined by the nature of the GT, but it may also depend on the value of the parameters and on the input.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Notes on the Graph Structures
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“However, implementing the operation this way may be very inefficient since the transducer can be very complicated.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Generalized Transduction
hedgedconditional- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
None of the sources cited for this claim have been checked yet.
In the paper
“The recognition of handwritten characters from a pen trajectory on a digitizing surface is often done in the time domain [44], [110], [111].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Preprocessing
What the sources say
- Not matchedcites [44] 44. P. Haffner and A. H. Waibel, "Time-delay neural networks embedding time alignment: A performance analysis," in Proc. EUROSPEECH'91, 2nd Europ. Conf. Speech Communication and Technology, Genova, Italy.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not checked yetcites [110] 110. S. Bengio and Y. Bengio, "An EM algorithm for asynchronous input/output hidden Markov models," in Proc. International Conference on Neural Information Processing, Hong-King, 1996, pp. 328-334.
- Not checked — no full text of the sourcecites [111] 111. C. Tappert, C. Suen, and T. Wakahara, "The state of the art in on-line handwriting recognition," IEEE Trans. Pattern Anal. Machine Intell., vol. 8, pp. 787-808, Dec. 1990. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Not matched
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The standard deviation of the weighted sum scales like the square root of the number of inputs when the inputs are independent, and it scales linearly with the number of inputs if the inputs are highly correlated.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > APPENDIX A PRECONDITIONS FOR FASTER CONVERGENCE
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
- The authors state this as an assumption: The standard deviation of the weighted sum scales linearly with the number of inputs if the inputs are highly correlated.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The beam search algorithm, however, implements pruning strategies which are appropriate for large interpretation graphs.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Network Training
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The Gauss-Newton approximation essentially ignores the nonlinearity of the estimated function (the NN, in our case), but not that of the loss function.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Additional information
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“The input-output HMM (IOHMM) [105], [109] is strongly related to GT's.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. GTN and HMM's
What the sources say
- Not checked yetcites [105] 105. Y. Bengio and P. Frasconi, "An input/output HMM architec- ture," in Advances in Neural Information Processing Systems, vol. 7, G. Tesauro, D. Touretzky, and T. Leen, Eds. Cam- bridge, MA: MIT, pp. 427-434, 1996.
- Not matchedcites [109] 109. M. Rahim, Y. Bengio, and Y. LeCun, "Discriminative feature and model design for automatic speech recognition," in Proc. Eurospeech '97, Rhodes, Greece.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
boostedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Convolutional NN's have been shown to eliminate the need for hand-crafted feature extractors.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Derivatives can be back propagated through the entire structure, although it is only practical to do it down to the segmenter.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Gradient-Based Learning
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Unfolding an HMM in time yields a graph that is very similar to our interpretation graph (at the final stage of processing of the GTN, before Viterbi recognition).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. GTN and HMM's
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The learning algorithms presented here are in a sense nothing more than unusual forms of gradient descent in complex, dynamic architectures, with efficient back-propagation algorithms to compute the gradient.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Unfortunately, using those derivatives leads to well-known problems associated with every Newton-like algorithm: these terms can be negative and can cause the gradient algorithm to move uphill instead of downhill.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Additional information
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Simultaneous automatic learning of segmentation and recognition can be achieved with gradient-based learning methods.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Graph composition using transducers (i.e., standard transduction) is easily and efficiently implemented as a generalized transduction.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Generalized Transduction
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Additional images were generated by randomly distorting the original images using simple affine transformations of the images.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Results
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
None of the sources cited for this claim have been checked yet.
In the paper
“Transducers have been applied to speech recognition [100] and language translation [101], and proposals have been made for handwriting recognition [102].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Previous Work
What the sources say
- Not checked — no full text of the sourcecites [100] 100. Y. Bengio, R. De Mori, G. Flammia, and R. Kompe, "Neural network-Gaussian mixture hybrid for speech recognition or density estimation," in Advances in Neural Information Process- ing Systems 4, J. E. Moody, S. J. Hanson, and R. P. Lippmann, Eds. Denver, CO: Morgan Kaufmann, 1992, pp. 175-182.
- Not checked yetcites [101] 101. F. C. N. Pereira and M. Riley, "Speech recognition by compo- sition of weighted finite automata," in Finite-State Devices for Natural Lague Processing. Cambridge, MA: MIT, 1997.
- Not checked — no full text of the sourcecites [102] 102. M. Mohri, "Finite-state transducers in language and speech processing," Computational Linguistics, vol. 23, no. 2, pp. 269-311, 1997.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This is the case in many audio signal recognition tasks, and visual scene analysis applications.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In addition, because those classifiers are much less invariant to scale variations than convolutional networks, it is necessary to multiply the number of scales at which the images are presented to the classifier.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Object Detection and Spotting with SDNN
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“All results are in writer independent mode (different writers in training and testing).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Experimental Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“During the short history of automatic pattern recognition, increasing the role of learning seems to have invariably improved the overall performance of recognition systems.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
None of the sources cited for this claim have been checked yet.
In the paper
“A different approach to graph-based trainable systems, called input-output HMM, was proposed in [104] and [105].”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Previous Work
What the sources say
- Not matchedcites [104] 104. M. Mohri and M. Riley, "Weighted determinization and mini- mization for large vocabulary recognition," in Proc. Eurospeech '97, Rhodes, Greece, pp. 131-134.
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not checked yetcites [105] 105. Y. Bengio and P. Frasconi, "An input/output HMM architec- ture," in Advances in Neural Information Processing Systems, vol. 7, G. Tesauro, D. Touretzky, and T. Leen, Eds. Cam- bridge, MA: MIT, pp. 427-434, 1996.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Not matched
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“But since those systems do not use convolutional networks, they cannot take advantage of the speedup described here, and they have to rely on other techniques, such as prefiltering and real-time tracking, to keep the computational requirement within reasonable limits.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Object Detection and Spotting with SDNN
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This relies on the probably incorrect assumption that the human designer is able to capture all the relevant information in the input.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The systems described in this paper are more evidence to this fact.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Even in the nontrainable parts of the system, the design philosophy in terms of graph transformation provides a clear separation between domain-specific heuristics (e.g., segmentation heuristics) and generic, procedural knowledge (the generalized transduction algorithm)”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > XI. CONCLUSIONS
boosted- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
In the paper
“Similar ideas have been applied to handwriting recognition (see [38] for a review).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Previous Work
What the sources say
- Not checked — no full text of the sourcecites [38] 38. J. Wang and J. Jean, "Multi-resolution neural networks for omnifont character recognition," in Proc. Int. Conf. Neural Networks, vol. III, 1993, pp. 1588-1593. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“If the initial weights are too small, the gradients are very small and the learning is slow.If they are too large, the sigmoids are saturated and the gradient is also very small.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > APPENDIX A PRECONDITIONS FOR FASTER CONVERGENCE
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The penalty function is differentiable, therefore its parameters are globally tunable.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. A GTN for Check Amount Recognition
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“In the stochastic version the gradient estimates are noisy, but the parameters are updated much more often than with the batch version.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Additional information
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Unlike most other second-order acceleration methods for back-propagation, the above method works in stochastic mode.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Additional information
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The bprop method is the basis of the back-propagation algorithm for generic GT's.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Notes on the Graph Structures
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“GTN's can be seen as a generalization and an extension of HMM's.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. GTN and HMM's
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In fact stochastic update must be better when there is redundancy, i.e., when a certain level of generalization is expected.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > Additional information
boosteddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In addition, it can be shown that the generalized transduction of a single graph is theoretically equivalent to the standard composition of this graph with a particular transducer graph.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Generalized Transduction
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Input normalization reduces intracharacter variability, thereby simplifying character recognition.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Preprocessing
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“It does not depend on stroke ordering or writing speed, and it can be used with all types of handwriting (capital, lower case, cursive, punctuation, symbols).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. Preprocessing
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Moreover, the absolute value of the second derivative of is a maximum at 1 and 1, which improves the convergence toward the end of the learning session.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > APPENDIX A PRECONDITIONS FOR FASTER CONVERGENCE
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“The recognizer was then inserted in the check-reading system and a small subset of the parameters were trained globally (at the field level) on whole check images.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“As before, using a differentiable function for computing the penalties will ensure that the parameters can be optimized globally.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. A GTN for Check Amount Recognition
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“A differentiable function is used to compute the data attached to the output arc from the data attached to the input arcs.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > A. A GTN for Check Amount Recognition
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“The network was then further trained on character images that had been automatically segmented from check images and manually truthed.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Results
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The only architectural parameters that remain to be selected are the number of feature maps in each layer and the information as to what feature map is connected to what other feature map.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > B. Network Architecture
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
What the paper did — it cites no source for this, so there is none to check. The quote shows it is in the paper.
In the paper
“Initial training on isolated characters was performed on a database of approximately 100 000 hand printed characters (95 classes of upper case, lower case, digits, and punctuation).”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Experimental Results
plainn = 95- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The reason for dividing by the fan-in is that we would like the initial standard deviation of the weighted sums to be in the same range for each unit and to fall within the normal operating region of the sigmoid.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > APPENDIX A PRECONDITIONS FOR FASTER CONVERGENCE
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This particular choice of parameters is merely a convenience, and does not affect the result.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > APPENDIX A PRECONDITIONS FOR FASTER CONVERGENCE
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds. It may state two separate claims as one.
The paper relies on this without support.
Real-life document recognition systems are composed of multiple modules including field extraction, segmentation, recognition, and language modeling.
“Real-life document recognition systems are composed of multiple modules including field extraction, segmentation, recognition, and language modeling.”
The paper relies on this without support.
The variability and richness of natural data make it almost impossible to build an accurate recognition system entirely by hand.
“Since the early days of pattern recognition it has been known that the variability and richness of natural data, be it speech, glyphs, or other types of patterns, make it almost impossible to build an accurate recognition system entirely by hand.”
The paper relies on this without support.
Most pattern recognition systems are built using a combination of automatic learning techniques and hand-crafted algorithms.
“Consequently, most pattern recognition systems are built using a combination of automatic learning techniques and hand-crafted algorithms.”
The paper relies on this without support.
The feature extractor contains most of the prior knowledge and is rather specific to the task.
“The feature extractor contains most of the prior knowledge and is rather specific to the task.”
The paper relies on this without support.
The classifier is often general purpose and trainable.
“The classifier, on the other hand, is often general purpose and trainable.”
The paper relies on this without support.
The recognition accuracy is largely determined by the ability of the designer to come up with an appropriate set of features.
“One of the main problems with this approach is that the recognition accuracy is largely determined by the ability of the designer to come up with an appropriate set of features.”
The paper relies on this without support.
A large proportion of modern commercial OCR systems use some form of multilayer NN trained with back propagation.
“As evidence of this fact, a large proportion of modern commercial OCR systems use some form of multilayer NN trained with back propagation.”
The paper relies on this without support.
The back-propagation procedure applied to multilayer NN's with sigmoidal units can solve complicated learning tasks.
“The third event was the demonstration that the back-propagation procedure applied to multilayer NN's with sigmoidal units can solve complicated learning tasks.”
The paper relies on this without support.
Most practical pattern recognition systems are composed of multiple modules.
“As stated earlier, most practical pattern recognition systems are composed of multiple modules.”
The paper relies on this without support.
Typically, each module is manually optimized, or sometimes trained, outside of its context.
“Typically, each module is manually optimized, or sometimes trained, outside of its context.”
The paper relies on this without support.
The input variables can be presented in any (fixed) order without affecting the outcome of the training.
“The input variables can be presented in any (fixed) order without affecting the outcome of the training.”
The paper relies on this without support.
Images have a strong 2-D local structure: variables that are spatially or temporally nearby are highly correlated.
“On the contrary, images (or time-frequency representations of speech) have a strong 2-D local structure: variables (or pixels) that are spatially or temporally nearby are highly correlated.”
The paper relies on this without support.
Local correlations are the reasons for the well-known advantages of extracting and combining local features before recognizing spatial or temporal objects.
“Local correlations are the reasons for the well-known advantages of extracting and combining local features before recognizing spatial or temporal objects, because configurations of neighboring variables can be classified into a small number of categories (e.g., edges, corners, etc.).”
The paper relies on this without support.
Drawing sensible conclusions from learning experiments requires that the result be independent of the choice of training set and test among the complete set of samples.
“Drawing sensible conclusions from learning experiments requires that the result be independent of the choice of training set and test among the complete set of samples.”
The paper relies on this without support.
Saturation of the sigmoids leads to slow convergence and ill-conditioning of the loss function.
“Saturation of the sigmoids must be avoided because it is known to lead to slow convergence and ill-conditioning of the loss function.”
The authors state this as an assumption.
The cut generation heuristic will include the correct set of cuts.
“The cut generation heuristic is designed so as to generate more cuts than necessary in the hope that the”
The authors state this as an assumption.
Penalties can be interpreted as negative log posteriors, so that posterior probabilities for an interpretation combine by summing and penalties combine as the negative logarithm of the sum of negative exponentials.
“We use a combination rule borrowed from a probabilistic interpretation of the penalties as negative log posteriors.”
The authors state this as an assumption.
A particular label sequence is given as the desired sequence to the GTN.
“In fact, assuming that a particular label sequence is given as the”
The paper relies on this without support.
No reliable segmentation heuristic exists for cursive handwriting.
“The SDNN architecture seems particularly attractive for recognizing cursive handwriting where no reliable segmentation heuristic exists.”
The paper relies on this without support.
With real training data, the correct sequence of labels for a string is generally available, but the precise locations of each corresponding character in the input image are unknown.
“With real training data, the correct sequence of labels for a string is generally available, but the precise locations of each corresponding character in the input image are unknown.”
The paper relies on this without support.
There has been no proposal for a systematic approach to multilayer graph-based trainable systems.
“However, there has been no proposal for a systematic approach to multilayer graph-based trainable systems.”
The authors state this as an assumption.
The function used by fprop to compute the values attached to its output arcs is differentiable.
“this method assumes that the function used by fprop to compute the values attached to its output arcs is differentiable.”
The paper relies on this without support.
Natural handwriting is often a mixture of different styles, i.e., lower case printed, upper case, and cursive.
“Natural handwriting is often a mixture of different “styles,” i.e., lower case printed, upper case, and cursive.”
The paper relies on this without support.
Characters taken in isolation can be very ambiguous, but considerable information is available from the context of the whole word.
“Characters taken in isolation can be very ambiguous, but considerable information is available from the context of the whole word.”
The paper relies on this without support.
The intent of the writer is to produce a legible image.
“Since the intent of the writer is to produce a legible image, it seems natural to preserve as much of the pictorial nature of the signal as possible, while at the same time exploit the sequential information in the trajectory.”
The paper relies on this without support.
Verifying the amount on a check is extremely time and money consuming for banks.
“The verification of the amount on a check is a task that is extremely time and money consuming for banks.”
The paper relies on this without support.
There is a very high interest in automating check verification as much as possible.
“As a consequence, there is a very high interest in automating the process as much as possible (see, for example, [112]- [114]).”
The paper relies on this without support.
Even a partial automation of check verification would result in considerable cost reductions.
“Even a partial automation would result in considerable cost reductions.”
The paper relies on this without support.
The threshold of economic viability for automatic check readers, as set by the bank, is when 50% of checks are read with less than 1% error and the other 50% are rejected.
“The threshold of economic viability for automatic check readers, as set by the bank, is when 50% of the checks are read with less than 1% error.The other 50% of the check being rejected and sent to human operators.”
The paper relies on this without support.
Checks contain at least two versions of the amount: the courtesy amount written with numerals and the legal amount written with letters.
“Checks contain at least two versions of the amount.The courtesy amount is written with numerals, while the legal amount is written with letters.”
The paper relies on this without support.
On business checks, amounts are generally machine-printed, relatively easy to read but quite difficult to find due to the lack of standard for business check layout.
“On business checks, which are generally machine-printed, these amounts are relatively easy to read but quite difficult to find due to the lack of standard for business check layout.”
The paper relies on this without support.
On personal checks, amounts are easy to find but much harder to read.
“On the other hand, these amounts on personal checks are easy to find but much harder to read.”
The paper relies on this without support.
Feature extraction is traditionally a fixed transform generally derived from expert prior knowledge about the task.
“Feature extraction is traditionally a fixed transform, and it is generally derived from some expert prior knowledge about the task.”
The paper relies on this without support.
Traditional recognition systems rely on many handcrafted heuristics to isolate individually recognizable objects.
“Traditional recognition systems rely on many handcrafted heuristics to isolate individually recognizable objects.”
The authors state this as an assumption.
The standard deviation of the weighted sum scales linearly with the number of inputs if the inputs are highly correlated.
“We chose to assume the second hypothesis since some units receive highly correlated signals.”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“An alternative would be the use of pooling ( LeCun et al., 1998 ), which however leads to loss of the exact position of the segmented voxel and thus can negatively impact accuracy.”
From Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation (2017) · cites this paper as LeCun et al., 1998
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“For image classification feed-forward deep networks, and in particular convolutional networks, are well-known to provide state-of-the-art results [26, 25].”
From Communication-Efficient Learning of Deep Networks from Decentralized Data (2016) · cites this paper as [26]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Overall, object detectors based on CNN [87] can be categorized into two primary types.”
From Object Detection in Autonomous Vehicles under Adverse Weather: A Review of Traditional and Deep Learning Approaches (2024) · cites this paper as [87]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“The normalization makes the network robust to illumination and contrast variation as demonstrated by [20].”
From Multimodal learning for facial expression recognition (2015) · cites this paper as [20]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“In the edge-assisted SLAM system, since video frames and the corresponding tiles are generated on the fly, lifelong learning enables the prediction model to be updated in an online way, thereby maintaining its prediction performance.”
From Orchestrating Joint Offloading and Scheduling for Low-Latency Edge SLAM (2025) · cites this paper as [27]
Experimental — accuracy not yet measured
The cited source says what the claim attributes to it.
In the citing paper
“Eigen et al. [46] proposed a system to remove dirt and raindrop noise from a single image, using the NN architecture originally developed by LeCun et al. [47].”
From A Review of Detection and Removal of Raindrops in Automotive Vision Systems (2021) · cites this paper as [47]
The cited source says what the claim attributes to it.
In the citing paper
“This method, which we describe in more detail in Supplementary Note 4, is highly efficient in terms of hardware usage and causes only moderate accuracy degradation compared to software-based neural networks on the two tasks considered here: Modified National Institute of Standards and Technology (MNIST) handwritten digit recognition30 and CIFAR-10 image recognition31.”
From Powering AI at the edge: A robust, memristor-based binarized neural network with near-memory computing and miniaturized solar cell (2024) · cites this paper as [30]
The cited source says what the claim attributes to it.
In the citing paper
“Moreover, even though there are reviews in this context [2], most of the surveyed works comprise Convolutional Neural Networks (CNN) [49] and Long short-term memory (LSTM) [32, 55], leaving behind some relevant and novel architectures, such as Autoencoders [108], Capsule Networks (CapsNet) [82], Deep Belief Networks [31], and Generative Adversarial Networks (GANs) [23].”
From Gait Recognition Based on Deep Learning: A Survey (2023) · cites this paper as [49]
The cited source says what the claim attributes to it.
In the citing paper
“In the context of gait recognition, Shiraga et al. [86] used an architecture similar to LENet [49] to create a gleaming-based recognizer through Gait Energy Image (GEI). The model achieved an accuracy of 91 . 5% in the OU-ISIR [59] Large Population dataset.”
From Gait Recognition Based on Deep Learning: A Survey (2023) · cites this paper as [49]
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“the NeuRRAM chip achieves 1.6-times to 2.3-times lower EDP and 7-times to 13-times higher computational density (measured by throughput per million of RRAMs) at various MVM input and output bit-precisions than previous state-of-the-art RRAM-based CIM chips, despite being fabricated at an older technology node17–27,36.”
From A compute-in-memory chip based on resistive random-access memory (2022) · cites this paper as [18]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Unlike the recentlyproposed Vision Transformer (ViT) that was designed for image classification specifically”
From Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions (2021) · cites this paper as [13]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“ViT has a columnar structure with coarse image patches as input.”
From Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions (2021) · cites this paper as [13]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Although ViT is applicable to image classification”
From Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions (2021) · cites this paper as [13]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“its output feature map is single-scale and low-resolution”
From Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions (2021) · cites this paper as [13]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“ViT (see Figure 1 (b)) is a columnar framework, whose output is single-scale.”
From Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions (2021) · cites this paper as [13]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Thus, it obtains a promising AP of 40.4 on COCO val2017 , 8.7 points higher than ViT-Small/32 (40.4 vs . 31.7).”
From Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions (2021) · cites this paper as [13]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“With increasing input scale, the growth rate of the GFLOPs of our PVT is greater than ResNet [22], but lower than ViT [13], as shown in Figure 6.”
From Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions (2021) · cites this paper as [13]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“Note that ViT and DeiT have limitations as they are specifically designed for classification tasks, and thus are not suitable for dense prediction tasks, which usually require effective feature pyramids.”
From Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions (2021) · cites this paper as [13]
Experimental — accuracy not yet measuredIn the citing paper
“Among different types of deep neural networks, DCNNs (LeCun et al. 1998(LeCun et al. , 2015;;Krizhevsky et al. 2012a) have brought about breakthroughs in processing images, video, speech and audio.”
From Deep Learning for Generic Object Detection: A Survey (2020) · cites this paper as LeCun et al. 1998
In the citing paper
“The CNN was first proposed and advocated for the high-dimensional image analysis by Lecun et al. 38”
From Artificial intelligence in healthcare: past, present and future (2017) · cites this paper as [38]
In the citing paper
“In the early days of modern deep learning one tended to use very simple combinations of the building blocks, as in Lenet [57] and AlexNet [1].”
From An overview of deep learning in medical imaging focusing on MRI (2019) · cites this paper as [57]
In the citing paper
“In medical imaging the interest in deep learning is mostly triggered by convolutional neural networks (CNNs) [57] 15 , a powerful way to learn useful representations of images and other structured data.”
From An overview of deep learning in medical imaging focusing on MRI (2019) · cites this paper as [57]
In the citing paper
“Recent studies have shown that deep learning techniques are successfully applied to medical domain.”
From Medical image retrieval using deep convolutional neural network (2017) · cites this paper as [40]
In the citing paper
“Convolutional Neural Networks (CNN) were introduced formally by Lecun, Bottou, Bengio, and Haffner (1998).They are inspired by the hierarchical process of simple and complex cells in the human brain to extract and learn different information from visual stimuli.”
From Multimodal emotional state recognition using sequence-dependent deep hierarchical features (2015) · cites this paper as Lecun, Bottou, Bengio, and Haffner (1998)
In the citing paper
“Since most of the studies are focused on image classification problems, most of the datasets in Table 7 are the datasets for image classification, such as MNIST (LeCun et al., 1998), CIFAR-10 ( Krizhevsky and Hinton, 2009), ImageNet (Fei-Fei, 2010), Fashion-MNIST (Xiao et al., 2017), SVHN (Netzer et al., 2011), and IMDB-WIKI dataset.”
From A software engineering perspective on engineering machine learning systems: State of the art and challenges (2021) · cites this paper as LeCun et al., 1998
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Rejecting Low Confidence Checks
“the discriminative forward loss function is an estimate of the logarithm of this probability.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > E. Remarks on Discriminative Training
“the exponential of minus can be interpreted as an estimate of the posterior probability of that label sequence given the input.”
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > D. Discriminative Forward Training
“the forward penalty of a graph is always larger than the forward penalty of a subgraph of this graph.”
§ Graph transformer network. HMM Hidden Markov model. HOS Heuristic oversegmentation > C. Forward Scoring and Forward Training
“The overall penalty will be smaller than all the penalties of the individual paths.”
Citations
Cites
The Nature of Statistical Learning Theory
Vapnik, V. N. (1995) · doi:10.1007/978-1-4757-2440-0 · Cited by 39,138
Wu, Y., & Vapnik, V. N. (1999) · Technometrics · doi:10.2307/1271368 · Cited by 27,236
A tutorial on hidden Markov models and selected applications in speech recognition
Rabiner, L. R. (1989) · Proceedings of the IEEE · doi:10.1109/5.18626 · Cited by 22,743
Numerical recipes in C: the art of scientific computing
(1993) · Choice Reviews Online · doi:10.5860/choice.30-5638 · Cited by 17,933
Learning Internal Representations by Error Propagation
Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1985) · doi:10.21236/ada164453 · Cited by 16,501
Advances in neural information processing systems 7
(1996) · Computers & Mathematics with Applications · doi:10.1016/s0898-1221(96)90463-0 · Cited by 14,759
Approximation by superpositions of a sigmoidal function
Cybenko, G. (1989) · Mathematics of Control, Signals, and Systems · doi:10.1007/bf02551274 · Cited by 13,883
Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex
Hubel, D. H., & Wiesel, T. N. (1962) · The Journal of Physiology · doi:10.1113/jphysiol.1962.sp006837 · Cited by 13,680
(1973) · CERN Document Server (European Organization for Nuclear Research) · Cited by 12,895
A training algorithm for optimal margin classifiers
Boser, B. E., Guyon, I. M., & Vapnik, V. N. (1992) · Proceedings of the fifth annual workshop on Computational learning theory · doi:10.1145/130385.130401 · Cited by 11,992
Research works printed but not linked: 27
- matan, 1992, Neural Information Processing Systems, 4
- lecun, 1987, Mod�les Connexionnistes de l'Apprentissage (Connectionist Learning Models)
- bottou, 1991, Advances in neural information processing systems, 3
- keeler, 1991, Neural Information Processing Systems, 3, 557
- pereira, 1997, Finite State Devices for Natural Language Processing
- ott, 1976, Construction of quadratic polynomial classifiers, Proc IEEE Int Conf Pattern Recognition, 161
- cortes, 1994, Advances in Neural Information Processing Systems 6, 327
- lippmann, 1987, Neural-net classifiers useful for speech recognition, Proc IEEE 1st Int Conf Neural Networks, 417
- driancourt, 1991, MLP, LVQ and DP: Comparison & cooperation, Proc Int Joint Conf Neural Networks, 2, 815
- haffner, 1992, Advances in neural information processing systems, 4, 579
- simard, 1993, Advances in neural information processing systems, 5
- lecun, 1997, Reading checks with graph transformer networks, Proc IEEE Int Conf Acoustics Speech Signal Processing, 1, 151
- bengio, 1996, An EM algorithm for asynchronous input/output hidden Markov models, Proc Int Conf Neural Information Processing, 328
- rahim, 0, Discriminative feature and model design for automatic speech recognition, Proc Eurospeech '97
- wolf, 1994, Advances in neural information processing systems, 6, 745
- mohri, 0, Weighted determinization and minimization for large vocabulary recognition, Proc Eurospeech '97, 131
- gilloux, 1993, Recognition of cursive script amounts on postal checks, Proc Europ Conf Postal Technol, 705
- bourlard, 1989, Advances in neural information processing systems, 1, 186
- bengio, 1992, Advances in neural information processing systems, 4, 175
- minsky, 1961, Learning in random nets, Proc 4th London Symp Information Theory, 335
- rumelhart, 1986, Parallel Distributed Processing Explorations in the Microstructure of Cognition, i, 318
- denker, 1995, The Mathematics of Induction
- lecun, 1985, A learning scheme for asymmetric threshold networks, Proc of COGNITIVA'85, 599
- kramer, 1988, Advances in neural information processing systems, 1, 40
- parker, 1985, Learning-logic
- bottou, 1991, Une approche th�orique de l'Apprentissage Connexionniste Applications � la reconnaissance de la Parole
- schenkel, 1993, Advances in neural information processing systems, 5, 723
Cited by
Sorted by citations
Items 57,801–57,900 of 61,544
Rotation-invariant representation learning by sector convolution neural networks
Lin, W., Sun, X., Zhong, C., Meng, H., Chen, G., Zhang, B., & Gu, Z. (2026) · Pattern Recognition · doi:10.1016/j.patcog.2026.113636
Workflow for Fine-Tuning and Evaluating DNA Language Models for Specific Genomics Issues
Nakamae, K., & Bono, H. (2026) · BIO-PROTOCOL · doi:10.21769/bioprotoc.5676
Interactive computer vision through the Web
Pizenberg, M. (2020) · theses.fr (ABES) · doi:10.70675/31d3db07z1123z48dczb7e2zd4e471c8fa5c
Kang, H. (2025) · IEEE Integrated STEM Education Conference (ISEC) · doi:10.1109/isec64801.2025.11460350
Deep learning onto graph space : application to image-based insect recognition
Martineau, C. (2019) · theses.fr (ABES) · doi:10.70675/96a7814czced9z4f31z9551zdf399ba8b52f
Productivité de l'agriculture française et volatilité des prix
Zheng, Y. (2018) · theses.fr (ABES) · doi:10.70675/28c4d27ezf654z45bcz91f2zc0cae650c5fa
Langlois, J. (2019) · theses.fr (ABES) · doi:10.70675/a7e2065cz9d46z492cz9f42z1c13fa04f4ee
Granger, N. (2019) · theses.fr (ABES) · doi:10.70675/9f7530bez3094z498fzbaf7zbfac62e25c00
Bio-inspired computing leveraging the synchronization of magnetic nano-oscillators
Talatchian, P. (2019) · theses.fr (ABES) · doi:10.70675/c925845bzbe12z482fz9b3bzffa8a1e65de0
High-resolution image deraining via dual-branch features interaction and fusion
Huang, W., Mei, J., Yan, T., Wang, Y., & Chang, X. (2026) · Neural Networks · doi:10.1016/j.neunet.2026.108936
Deep learning and featured-based classification techniques for radar imagery
Belloni, C. (2019) · theses.fr (ABES) · doi:10.70675/f03df957z1747z4ac6zaf13zb02051887bd0
Contribution to graph-based manifold learning with application to image categorization
Zhu, R. (2020) · theses.fr (ABES) · doi:10.70675/92e59429za01cz478azb581zdd5659a28bde
Detecting AI-generated videos via global semantic awareness and inter-frame semantic consistency
Xiang, Q., Guo, Y., Gu, S., Xie, L., & Yining, H. U. (2026) · Neurocomputing · doi:10.1016/j.neucom.2026.133530
Toward unsupervised human activity and gesture recognition in videos
Negin, F. (2018) · theses.fr (ABES) · doi:10.70675/cbc55a84zee02z4838zbafczf15751f62c2f
Input noise injection for supervised machine learning, with applications on genomic and image data
Khalfaoui, B. (2019) · theses.fr (ABES) · doi:10.70675/2026ab7fzf56cz4bd4z85daz4cbfb2fbc88a
Sémantisation à la volée de nuages de points 3D acquis par systèmes embarqués
Roynard, X. (2019) · theses.fr (ABES) · doi:10.70675/8f872df1z141bz4874zbdedz8be0859ef26d
du Terrail, J. O. (2018) · theses.fr (ABES) · doi:10.70675/083d206az0a4ez4480z866ez53e9de431d15
Modélisation des séries temporelles par apprentissage profond
Haykal, V. (2019) · theses.fr (ABES) · doi:10.70675/5f7a222bzd740z48a2z9951z429c68aa91a3
Unsupervised cross-lingual representation modeling for variable length phrases
Liu, J. (2020) · theses.fr (ABES) · doi:10.70675/d93d53a6z857fz47e6z8dc8z85289e135ca6
Détection et analyse des évènements rares par vision, dans un contexte urbain ou péri-urbain
Atrevi, D. F. (2019) · theses.fr (ABES) · doi:10.70675/45ab3162z1f59z4314zac77z515f812fa17e
Capellier, É. (2020) · theses.fr (ABES) · doi:10.70675/654a5182z28e7z454bz83d5z6a37dfbc4fc2
Brain-inspired computing leveraging the transient non-linear dynamics of magnetic nano-oscillators
Riou, M. (2019) · theses.fr (ABES) · doi:10.70675/8aa2d912z0594z46e3zb96czfd1e1c3034ff
La restructuration des documents graphiques destructurés
Péré-Laperne, J. (2019) · theses.fr (ABES) · doi:10.70675/8a9c38e9zc387z49fcz9371z57c005ba3d3d
Facial landmark detection with local and global motion modeling
Belmonte, R. (2019) · theses.fr (ABES) · doi:10.70675/20a5387az1ee3z4b59zad56z39522e0e4083
Modélisation géométrique personalisée du membre inférieur à partir de radiographies bi-planes
Girinon, F. (2018) · theses.fr (ABES) · doi:10.70675/af864f70z8fefz4ce3z9b7fz5578c83d937e
Geometric operators for 3D modeling using dictionary-based shape representations
Lescoat, T. (2020) · theses.fr (ABES) · doi:10.70675/40c3343fz789dz4faaz856bz98c1c937bb68
Contribution au développement de l’apprentissage profond dans les systèmes distribués
Hardy, C. (2019) · theses.fr (ABES) · doi:10.70675/e16a405fzdd4ez4f0eza62dzd5e22a39af53
Deep Depth from Defocus : Neural Networks for Monocular Depth Estimation
Carvalho, M. (2019) · theses.fr (ABES) · doi:10.70675/4c4df4cbz477az42b5zb2e2z0a8ec821bf74
Computational detection of socioeconomic inequalities
Abitbol, J. L. (2020) · theses.fr (ABES) · doi:10.70675/ec0a1334z0ceez4affzb5f9z1cfcff5ecbba
Classification de données massives de télédétection
Audebert, N. (2018) · theses.fr (ABES) · doi:10.70675/e7e6756dzea11z49c1z99dfz2a794a46b421
Video inpainting and semi-supervised object removal
Le, T. T. (2019) · theses.fr (ABES) · doi:10.70675/1cf4c899z91b9z479czafcdz6dce7a3b070a
Exploiting imprecise information sources in sequential decision making problems under uncertainty
Drougard, N. (2015) · theses.fr (ABES) · doi:10.70675/20fdeab0z13ecz473az9312zf7b5934da3b4
Design of optimization algorithms for large scale continuous problems : application on deep learning
Souquet, L. (2019) · theses.fr (ABES) · doi:10.70675/aeaa8997z5886z45f9zbd76z62b32aaf2036
Learning Image-to-Surface Correspondence
Güler, R. A. (2019) · theses.fr (ABES) · doi:10.70675/5c92c11bzafbaz442fz9accz878a6d7dbaf1
Bayesian statistical inference and deep learning for primordial cosmology and cosmic acceleration
Ramanah, D. K. (2019) · theses.fr (ABES) · doi:10.70675/454bb359z4ed4z4eebzbc32zed2ada8748c5
Intégration de Savoir Expert dans les Modèles Génératifs : Application à la Génération Musicale
Rivaud, S. (2020) · theses.fr (ABES) · doi:10.70675/ebde687cz5636z4e96z87c1z1d743f18dbd3
Neural architectures for estimating correspondences between images
Rocco, I. (2020) · theses.fr (ABES) · doi:10.70675/fe35f3e3z9b1fz4f38z9dfcz9f789e7e4139
Data driven methods to support decision making in Deep Brain Stimulation for Parkinson’s disease
Peralta, M. (2020) · theses.fr (ABES) · doi:10.70675/59262743z5225z46d3z984ez3f9b0a19c4a7
Deep Neural Networks for the segmentation and classification in Medical Imaging
Naceur, M. B. (2020) · theses.fr (ABES) · doi:10.70675/17ba8dc8z1aabz4f52z9ebazb6af0a940b08
Neural Methods for Sentiment Analysis and Text Summarization
Le, T.-H. (2020) · theses.fr (ABES) · doi:10.70675/e5daf161ze829z4e0fzbd75zb727195984b2
Modèles neuronaux pour la représentation et l'appariement d'objets géotextuels
Mousset, P. (2020) · theses.fr (ABES) · doi:10.70675/c533c1f4z5674z4c7cz93e9ze76be342daaa
Deep generative models : over-generalisation and mode-dropping
Lucas, T. (2020) · theses.fr (ABES) · doi:10.70675/8b7d62c3z920dz4157z9861z68bd3a8f62ee
Learning with weak supervision using deep generative networks
Chen, M. (2020) · theses.fr (ABES) · doi:10.70675/c36a0451za9e0z4328z9126z293a288f0fda
Machine learning over spaces of measures : invariant deep networks and quantile regression
de Bie, G. (2020) · theses.fr (ABES) · doi:10.70675/ba087ccbz1de2z4383z84c2z239659c7ad14
Hardware Accelerators for Machine Learning Applications. Case Study : Autonomous Vehicles
Neggaz, M. A. (2020) · theses.fr (ABES) · doi:10.70675/09cf40c0z17a0z4efbz8c0fzb956e149d59c
Détection des adventices par imagerie aérienne
Bah, M. D. (2020) · theses.fr (ABES) · doi:10.70675/1b2b47f5z2ea5z45aez891fz2240f70d124d
Efficient Representations of Dynamic Textures
Nguyen, T. T. (2020) · theses.fr (ABES) · doi:10.70675/2bb11ef1zf8caz4528zb4fez2c9455908822
Spatial techniques for video understanding
Sydorov, V. (2021) · theses.fr (ABES) · doi:10.70675/ab9825adz1fb8z4f1cz9217zef67d5530022
Deep kernel representation learning for complex data and reliability issues
Laforgue, P. (2020) · theses.fr (ABES) · doi:10.70675/e8c2e6ccz9ff4z451cza8a2zf31fdf4eabd5
Crop Mapping Using Remote Sensing Data and Deep Learning Approaches
Hai, V. Q., Truong, N. X., Khoi, L. N., & Thanh, N. T. N. (2026) · Communications in computer and information science · doi:10.1007/978-3-032-21625-0_34
De la navigation visuelle à l’analyse sémantique pour véhicules autonomes
Hrustic, E. (2021) · theses.fr (ABES) · doi:10.70675/eff83957zccd2z4ba5zb7a7z8796ee0320f3
Analyse automatique d'images aériennes historiques : application à une étude épidémiologique
Ratajczak, R. (2020) · theses.fr (ABES) · doi:10.70675/1da5d422zffd7z4f92z8677z7a6c8fcb81ef
Unsupervised and weakly supervised deep learning methods for computer vision and medical imaging
Sahasrabudhe, M. (2020) · theses.fr (ABES) · doi:10.70675/053d6ac3z8c79z451cz88c8z553669be02c2
Applications of machine learning to the resolution of recurrent combinatorial optimization problems
Mossina, L. (2020) · theses.fr (ABES) · doi:10.70675/69eb2c60z998cz4ff6zaef7z4d002022bc97
Resistive memory co-design in CMOS technologies
Bazzi, H. (2020) · theses.fr (ABES) · doi:10.70675/78f0c866z7434z4bd7z84b7zb84b151fd845
Cancer Detection in Full Field Optical Coherence Tomography Images
Mandache, D. (2022) · theses.fr (ABES) · doi:10.70675/a8e12287zf665z4a1fz822fze7174f880867
Deep Face Analysis for Aesthetic Augmented Reality Applications
Yan, Y. (2020) · theses.fr (ABES) · doi:10.70675/407a158cze060z45abz9868zbd9a5ea5b2ee
Analyse des sentiments et des émotions de commentaires complexes en langue française
Pecòre, S. (2019) · theses.fr (ABES) · doi:10.70675/cb268a2cz7f02z4d0bz81f4zf4ddfe40c7f9
Weakly supervised learning for image classification and potentially moving obstacles analysis
Chiaroni, F. (2020) · theses.fr (ABES) · doi:10.70675/7280ff3cz9ff6z48b5z96a9zc27c3374f45e
Multimodal and interactive models for visually grounded language learning
Strub, F. (2020) · theses.fr (ABES) · doi:10.70675/565f6b44z5446z433azb85dz57d7ca36bccf
Solar wind / magnetosphere coupling inferred from machine-learning methods
Nguyen, G. (2021) · theses.fr (ABES) · doi:10.70675/6bd46f10z9861z44f3z82bfz3fea6c7511ed
Robust image representation for classification, retrieval and object discovery
Siméoni, O. (2020) · theses.fr (ABES) · doi:10.70675/6333f15fz11eaz4848z9ed3z17978d8e10b2
Approximations parcimonieuses et méthodes à noyaux pour la compression de modèles d'apprentissage
Giffon, L. (2020) · theses.fr (ABES) · doi:10.70675/7c477a6fz884bz4705z871fz046d746e8042
Study of machine learning methods for optimization and reliability improvements of high power linacs
Debongnie, M. (2021) · theses.fr (ABES) · doi:10.70675/3ffa0daaz1b55z4236z8c99z1855de370c84
Solving jigsaw puzzles with deep learning
Paumard, M.-M. (2020) · theses.fr (ABES) · doi:10.70675/89177e02z5cedz4971za308z1f26dac68995
Digital Implementation of Neuromorphic systems using Emerging Memory devices
Hirtzlin, T. (2020) · theses.fr (ABES) · doi:10.70675/7c42d7b9z4001z4424z8a0az6c7cb2079ec8
Méta-apprentissage par renforcement pour le contrôle adaptatif
Jaafra, Y. (2020) · theses.fr (ABES) · doi:10.70675/4a9b4240zabb1z45abz8446z5a014f63e1c0
Automatic learning of next generation human-computer interactions
Debard, Q. (2020) · theses.fr (ABES) · doi:10.70675/8aa26513zedabz4492zbfacz4c4613e74be6
From Text to Faces: A Systematic Review of Stable Diffusion in Facial Image GenerationPreprint
Souza, M. A., Oliveira, B. D. B., & Vidal, F. B. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6521443
Flexible clustering algorithms for heterogeneous datasets
Roizman, V. (2021) · theses.fr (ABES) · doi:10.70675/5e77d24dz77d1z47cczb786z556525a66da1
Estimation de poses 3D multi-personnes à partir d'images RGB
Benzine, A. (2020) · theses.fr (ABES) · doi:10.70675/2b8b4e1bzb956z4f5cz8655zc7dbff1fdbf9
Accelerating Evolutionary Design Exploration with Predictive and Generative Models
Gaier, A. (2020) · theses.fr (ABES) · doi:10.70675/6c59cea5zff2cz47b4z9de3z0a17f6328548
Robust detection of astronomical sources using convolutional neural networks
Paillassa, M. (2020) · theses.fr (ABES) · doi:10.70675/eadcd5fdz3c39z4c82z9901z314800455d3d
Evolving principes of artificial neural design
Wilson, D. G. (2019) · theses.fr (ABES) · doi:10.70675/b678f35dz7ee5z441ez93f3ze801472a9a28
Thiry, L. (2021) · theses.fr (ABES) · doi:10.70675/0f9f0c35z84e2z46e6zbb1cz76600e309fdf
Semi-supervised dictionary learning and Semi-supervised deep neural network
Tran, K.-H. (2021) · theses.fr (ABES) · doi:10.70675/42000ac0z65ebz4a00z9dd3z6af0c02d8611
Analysis and synthesis of urban sound scenes using deep learning techniques
Gontier, F. (2020) · theses.fr (ABES) · doi:10.70675/b3ca4067zb65az44bdzb851z6f627675aa58
Content and data linking leveraging ontological knowledge in data journalism
Vaigh, C. B. E. (2021) · theses.fr (ABES) · doi:10.70675/6d3a4313z8d77z4dddz8cfcz3e88306088d4
Dynamic visual saliency in image sequences
Maczyta, L. (2020) · theses.fr (ABES) · doi:10.70675/44a50fa3zc71fz4d80za512zcc3021e0fbee
Diagnostic automatisé des pathologies de la rétine à l'aide des volumes OCT
Safarjalani, R. (2020) · theses.fr (ABES) · doi:10.70675/9fab8ce5z31e1z4fd0z86aez2ab2f205a18a
Metadata
- Authors
- Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P.
- Affiliations
- AT&T (United States) (US)
- Université de Montréal (CA)
- Venue
- Proceedings of the IEEE
- Published
- 1998
- Volume
- 86
- Issue
- 11
- Pages
- 2278-2324
- ISSN
- 0018-9219 (ISSN-L)
- OpenAlex
- W2112796928
- Semantic Scholar
- 162d958ff885f1462aeda91cd72582323fd6a1f4
- Cited by
- 58,805
- References
- 129
- Citations by year
- 2026: 3932 · 2025: 6037 · 2024: 5859 · 2023: 5924 · 2022: 5751 · 2021: 6779 · 2020: 6359 · 2019: 5708 · 2018: 4169 · 2017: 2778 · 2016: 1818 · 2015: 1223 · 2014: 644 · 2013: 396 · 2012: 267
- FWCI
- 12.51
- Type
- journal article
- Language
- en
- Open access
- green
- hal.science
- Locations
- Proceedings of the IEEE (publishedVersion)
- HAL (submittedVersion, other-oa)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Handwritten Text Recognition Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Vehicle License Plate Recognition
- Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Applications
- Keywords
- document recognition, handwritten character recognition, handwritten digit recognition, convolutional neural networks, graph transformer networks, global training, backpropagation, online handwriting recognition, field extraction, language modeling
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-11-08, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-11-08 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
