The DOI matches its Crossref record; 142 of the 149 references it cites were found in a registry or an index; 1 of the 84 sources its claims cite were checked for support.
Deep learning for time series classification: a review
Open accessOverview
AI snapshot
Written from the paper’s full text.
TL;DR
This review empirically compares nine end-to-end deep learning architectures for time series classification across 97 univariate and multivariate datasets, training 8,730 models. It finds that a deep residual network (ResNet) is the strongest DNN, significantly outperforming most other deep models and reaching performance not significantly different from COTE/HIVE-COTE on the UCR/UEA archive.
Key findings
- ResNet significantly outperformed the other deep learning approaches on the 85 univariate UCR/UEA datasets, with an average rank of almost 2, winning on 50 out of 85 problems; FCN and Encoder followed as second and third.
- ResNet was not significantly different from COTE/HIVE-COTE in pairwise statistical tests, while NN-DTW-WW and EE showed the lowest average ranks among compared state-of-the-art methods.
- MCNN and t-LeNet, which use window slicing/subsequence extraction, yielded very low accuracies with only one win (Earthquakes); MCDCNN and Time-CNN performed similarly to MLP.
- ResNet was more stable across random initializations than FCN, with average standard deviation 1.48 versus 1.70; on DiatomSizeReduction, ResNet and FCN reached only 30% accuracy while Time-CNN reached 95%.
Methods
- The authors reimplemented nine discriminative end-to-end deep learning models (MLP, FCN, ResNet, Encoder, MCNN, t-LeNet, MCDCNN, Time-CNN, TWIESN) in a common Keras/TensorFlow framework and trained each with 10 random initializations on 97 datasets: 85 univariate UCR/UEA datasets and 12 multivariate datasets.
- In total, 8,730 experiments were run on a cluster of 60 GPUs; accuracy was averaged over the 10 runs, and comparisons with published state-of-the-art results used median test error.
- Statistical comparison used the Friedman test followed by pairwise Wilcoxon signed-rank tests with Holm’s alpha correction (5%), visualized with critical difference diagrams.
Limitations
- The paper states that generative models and models using hand-engineered features were excluded, and that transfer learning, data augmentation, training/testing time, and normalization effects were not thoroughly studied.
- The paper acknowledges that non-deep state-of-the-art algorithms were not trained or tuned by the authors, so the comparison may favor DNNs; it also notes that the compared architectures may have been overfitted or empirically designed for the UCR/UEA archive.
- Evident from the text: the multivariate archive contains only 12 datasets, and the statistical test failed to find significant differences among the nine classifiers on that archive; ResNet’s hyperparameters were not tuned per dataset, and random initialization can significantly affect DNN accuracy.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
63 Not checked yet · 47 Own interpretation · 44 Own result · 22 No basis given · 6 Own method · 1 Supported so far
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“Hills et al. (2014) introduced a shapelet transformation based approach that discovered shapelets that are similar to the ones identified by Ye and Keogh (2011).”
What the sources say
- Not checked yetcites Hills et al. (2014) source
- Supportedcites Ye and Keogh (2011) source
The citation to Ye and Keogh (2011) identifies the earlier shapelets, rather than sourcing the description of Hills et al.’s later approach. Section 5.1.1 says Ye and Keogh reported shapelets from their experiments; Section 3 describes how they found them. The paper is not being cited as evidence of a comparison with the later work. (LLM, conf=0.88)
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.
None of the sources cited for this claim have been checked yet.
In the paper
“our experiments showed that not only DNNs are able to significantly outperform the NN-DTW, but are also able to achieve results that are not significantly different than COTE and HIVE-COTE using a deep residual network architecture (He et al., 2016; Wang et al., 2017b).”
What the sources say
- Not checked yetcites He et al., 2016
- Not checked yetcites Wang et al., 2017b
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.
- Rests on an assumption: The UCR/UEA archive datasets are representative of TSC problems.
- Rests on an assumption: The deep learning implementations in the framework faithfully represent the original methods.
None of the sources cited for this claim have been checked yet.
In the paper
“our experiments showed that not only DNNs are able to significantly outperform the NN-DTW, but are also able to achieve results that are not significantly different than COTE and HIVE-COTE using a deep residual network architecture (He et al., 2016; Wang et al., 2017b).”
What the sources say
- Not checked yetcites He et al., 2016
- Not checked yetcites Wang et al., 2017b
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.
- Rests on an assumption: The UCR/UEA archive datasets are representative of TSC problems.
- Rests on an assumption: The deep learning implementations in the framework faithfully represent the original methods.
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 addition, one of the advantages of FCNs is the invariance (denoted by # invar in Table 1) in the number of parameters for 4 layers (out of 5) across time series of different lengths.”
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
“We should note that twenty datasets contains a relatively small training set (50 or fewer instances) which surprisingly was not an impediment for obtaining high accuracy when applying a very deep architecture such as ResNet.”
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
“During our study, we found that CNN is the most widely applied architecture for the TSC problem”
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
“In our experiments, when comparing to state-of-the-art algorithms, we have encountered this problem with (ResNet > COTE > HIVE- COTE).”
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
“where the statistical test failed to find any significant difference between the nine classifiers”
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
“and significantly outperforms the FCN architecture.”
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
“One exception is the electrocardiography (ECG) datasets (7 in total) where ResNet was drastically beaten by the FCN model in 71.4% of ECG datasets.”
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.
- Probably not a substantial claim. 47
- It may state two separate claims as one. 52
- The quote only partly states this claim. 2
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
“Finally, we analyze how poor random initializations can have a significant effect on a DNN's performance.”
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
“We can see that FCN does not hold any pooling nor a regularization operation.”
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 to the FCN model, the layers (except the final one) in the ResNet architecture have an invariant number of parameters across different datasets.”
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 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 ResNet significantly outperforms the other approaches with an average rank of almost 2.”
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
“This suggestion is also supported by the fact that Time-CNN is the smallest model: it contains a very small number of parameters by design with only 18 filters compared to the 512 filters of FCN.”
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
“Again, ResNet and FCN still dominate with not much of a difference.”
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.
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
“our empirical study strongly suggests to use ResNet instead of any other deep learning algorithm - it is the most accurate one with similar runtime to FCN (the second most accurate DNN).”
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.
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
“Although the datasets used in our experiments are relatively small compared to the billions of labeled images (such as ImageNet (Russakovsky et al., 2015) and OpenImages (Krasin et al., 2017) challenges), the deepest networks did reach competitive accuracies on the UCR/UEA archive benchmark.”
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 MCNN and t-LeNet architectures yielded very low accuracies with only one win for the Earthquakes dataset.”
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
“FCN wins on 36 datasets whereases Encoder wins only on 17”
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
“ResNet and FCN achieved the worst accuracy (30%) on this dataset while Time-CNN reached the best accuracy (95%).”
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
“However, TWIESN did reach competitive accuracies on several long time series datasets such as reaching a 96.8% accuracy on Meat whose time series length is equal to 448.”
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
“Overall, our results show that, on average, ResNet is the best architecture with FCN and Encoder following as second and third respectively.”
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
“In other words, FCN's learned representation has too many data points close to the decision boundary whereas ResNet's hidden features enables projecting data points further away from the decision boundary.”
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.
- Rests on an assumption: MDS plots with Euclidean distance on GAP vectors preserve class separability sufficiently to interpret model performance.
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
“which is evident by the relatively low accuracy of both models FCN and ResNet which is equal respectively to 58.7% and 74.4%.”
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
“In other words, FCN's CAM contains much more smoother regions with cyan, green and yellow regions, whereas ResNet's CAM contains more dark red and blue subsequences”
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
“ResNet performed very well in general except for the ECG datasets where it was outperformed by FCN.”
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
“In other words, the weight's initial value can easily decrease the accuracy of FCN whereas ResNet maintained a relatively high accuracy when taking the worst initial weight values.”
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
“We found small variance between the approaches that replace the GAP layer with an FC dense layer (MCDCNN, CNN) which also showed similar performance to TWIESN and MLP.”
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
“We note that both deep learning models (FCN and ResNet) managed to project the data from GunPoint into a linearly separable space”
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.
- Rests on an assumption: MDS plots with Euclidean distance on GAP vectors preserve class separability sufficiently to interpret model performance.
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
“we observed in our experiments, that DNNs for TSC suffer from a significant decrease (increase) in accuracy when initialized with bad (good) random weights.”
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.
In the paper
“DNNs have indeed revolutionized the field of computer vision especially with the advent of novel deeper architectures such as Residual and Convolutional Neural Networks.”
- Also stated in: § 1 Introduction · p. 2 · plain
“Deep Convolutional Neural Networks (CNNs) have revolutionized the field of computer vision (Krizhevsky et al., 2012).”
What the sources say
- Not checked yetcites Krizhevsky et al., 2012
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. It may state two separate claims as one.
- Also stated in: § 1 Introduction · p. 2 · plain
None of the sources cited for this claim have been checked yet.
In the paper
“To achieve its high accuracy, HIVE-COTE becomes hugely computationally intensive and impractical to run on a real big data mining problem (Bagnall et al., 2017).”
- Also stated in: § 1 Introduction · p. 2 · hedged
“The approach requires training 37 classifiers as well as cross-validating each hyperparameter of these algorithms, which makes the approach infeasible to train in some situations (Lucas et al., 2018).”
What the sources say
- Not checked yetcites Bagnall et al., 2017 source
- Not checked yetcites Lucas et al., 2018
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: § 1 Introduction · p. 2 · hedged
None of the sources cited for this claim have been checked yet.
In the paper
“Apart from images, sequential data such as text and audio can also be processed with DNNs to reach state-of-the-art performance for document classification and speech recognition.”
- Also stated in: § 1 Introduction · p. 2 · plain
“Following the success of deep neural networks (DNNs) in computer vision, a huge amount of research proposed several DNN architectures to solve natural language processing (NLP) tasks such as machine translation (Sutskever et al., 2014; Bahdanau et al., 2015), learning word embeddings (Mikolov et al., 2013; Mikolov et al., 2013) and document classification (Le and Mikolov, 2014; Goldberg, 2016). DN”
What the sources say
- Not checked yetcites Sutskever et al., 2014
- Not checked yetcites Hinton et al., 2012 source
- Not checked yetcites Le and Mikolov, 2014
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.
- Also stated in: § 1 Introduction · p. 2 · plain
None of the sources cited for this claim have been checked yet.
In the paper
“Once we have established that a statistical difference exists within the classifiers' performance, we followed the pairwise post-hoc analysis recommended by Benavoli et al. (2016) where the average rank comparison is replaced by a Wilcoxon signed-rank test (Wilcoxon, 1945) with Holm's alpha correction (Holm, 1979; Garcia and Herrera, 2008).”
What the sources say
- Not checked yetcites Benavoli et al., 2016
- Not checked yetcites Wilcoxon, 1945 source
- Not checked yetcites Holm, 1979
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 multivariate evaluation shows another benefit of deep learning models, which is the ability to handle the curse of dimensionality (Bellman, 2010; Keogh and Mueen, 2017) by leveraging different degrees of smoothness in compositional function (Poggio et al., 2017) as well as the parallel computations of the GPUs (Lu et al., 2015).”
What the sources say
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.
- Rests on an assumption: The UCR/UEA archive datasets are representative of TSC problems.
In the paper
“Apart from time series forecasting, we found that these neural networks were rarely applied for time series classification which is mainly due to three factors: (1) the type of this architecture is designed mainly to predict an output for each element (time stamp) in the time series (L¨ angkvist et al., 2014);”
What the sources say
- Not checked yetcites L¨ angkvist et al., 2014
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
“(3) RNNs are considered hard to train and parallelize which led the researchers to avoid using them for computational reasons (Pascanu et al., 2013).”
What the sources say
- Not checked yetcites Pascanu et al., 2013
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.
In the paper
“They were designed to mitigate the challenges of RNNs by eliminating the need to compute the gradient for the hidden layers which reduces the training time of these neural networks thus avoiding the vanishing gradient problem.”
What the sources say
- Not checked yetcites Gallicchio and Micheli, 2017
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.
None of the sources cited for this claim have been checked yet.
In the paper
“This is surprising as deep learning has seen very successful applications in the last years.”
- Also stated in: § 1 Introduction · p. 2 · plain
“After having established the current state-of-the-art of non deep classifiers for TSC (Bagnall et al., 2017), we discuss the success of Deep Learning (LeCun et al., 2015) in various classification tasks which motivated the recent utilization of deep learning models for TSC (Wang et al., 2017b).”
What the sources say
- Not checked yetcites LeCun et al., 2015 source
- Not checked yetcites Wang et al., 2017b
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.
- Also stated in: § 1 Introduction · p. 2 · plain
In the paper
“Deep learning approaches for TSC can be separated into two main categories: the generative and the discriminative models (as proposed in L¨ angkvist et al. (2014)).”
What the sources say
- Not checked yetcites L¨ angkvist et al., 2014
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
“As for comparing the classifiers over multiple datasets, we followed the recommendations in Demˇ sar (2006) and used the Friedman test (Friedman, 1940) to reject the null hypothesis.”
What the sources say
- Not checked yetcites Demˇ sar, 2006
- Not checked yetcites Friedman, 1940 source
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.
In the paper
“Generative models usually exhibit an unsupervised training step that precedes the learning phase of the classifier (L¨ angkvist et al., 2014).”
What the sources say
- Not checked yetcites L¨ angkvist et al., 2014
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
“Finally, adding to the huge runtime of HIVE-COTE, the decision taken by 37 classifiers cannot be interpreted easily by domain experts, since researchers already struggle with understanding the decisions taken by an individual classifier.”
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
“MLP does not have any layer whose number of parameters is invariant across time series of different lengths (denoted by # invar in Table 1) which means that the transferability of the network is not trivial: the number of parameters (weights) of the network depends directly on the length of the input time series.”
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
“One final note is that the careful design of Encoder's attention mechanism enabled the invariance across all layers which encouraged the authors to implement a transfer learning approach.”
What the sources say
- Not checked yetcites Serr` a et al., 2018
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.
In the paper
“Residual Networks (ResNets) (Wang et al., 2017b; Geng and Luo, 2018) which add linear shortcut connections for the convolutional layers potentially enhancing the model's accuracy (He et al., 2016).”
What the sources say
- Not checked yetcites He et al., 2016
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.
In the paper
“To measure the performance of the model on the test data (generalization), we adopted the accuracy measure (similarly to Bagnall et al. (2017)).”
What the sources say
- Not checked yetcites Bagnall et al., 2017 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.
In the paper
“Although ESNs were originally proposed for time series forecasting, Tanisaro and Heidemann (2016) proposed a variant of ESNs that uses directly the raw input time series and predicts a probability distribution over the class variables.”
What the sources say
- Not checked yetcites Tanisaro and Heidemann, 2016
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.
In the paper
“In addition, one of the main characteristics of this architecture is the replacement of the traditional final FC layer with a Global Average Pooling (GAP) layer which reduces drastically the number of parameters in a neural network while enabling the use of the CAM (Zhou et al., 2016) that highlights which parts of the input time series contributed the most to a certain classification.”
What the sources say
- Not checked yetcites Zhou et al., 2016
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
“The latter approach reached higher accuracy thus benefiting from the transfer learning technique.”
What the sources say
- Not checked yetcites Serr` a et al., 2018
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
“In contrast to feature engineering, end-to-end deep learning aims to incorporate the feature learning process while fine-tuning the discriminative classifier (Nweke et al., 2018).”
What the sources say
- Not checked yetcites Nweke et al., 2018 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
“which is probably due to their robustness and the relatively small amount of training time compared to complex architectures such as RNNs or MLPs.”
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.
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
“For memory usage limitations over a single GPU, we left the MTS dataset Performance Measurement System (PeMS) out of our experimentations.”
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
“Unlike ResNet and FCN, this approach does not have much invariant layers (2 out of 4) due to the use of an FC layer instead of a GAP layer, thus increasing drastically the number of parameters needed to be trained which also depends on the length of the input time series.”
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
“This has also been the consensus in the human activity recognition literature, where the accuracy of deep learning methods depends highly on the quality of the extracted features (Nweke et al., 2018).”
What the sources say
- Not checked yetcites Nweke et al., 2018 source
boostedcorrelational- 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
“We also evaluated all deep learning models on Baydogan's archive (Baydogan, 2015) that contains 13 MTS classification datasets.”
What the sources say
- Not checked yetcites Baydogan, 2015
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
“However, any deep learning algorithm can benefit from this type of modification.”
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
“That being said, we can easily pre-train a model on a source dataset, then transfer and fine-tune it on a target dataset without having to modify the hidden layers of the network.”
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
“This archive also exhibits datasets with different characteristics such as the length of the time series which, unlike the UCR/UEA archive, varies among the same dataset. This is due to the fact that the datasets in the UCR/UEA archive are already re-scaled to have an equal length among one dataset (Bagnall et al., 2017).”
What the sources say
- Not checked yetcites Bagnall et al., 2017 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. 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
“FCNs are mainly convolutional networks that do not contain any local pooling layers which means that the length of a time series is kept unchanged throughout the convolutions.”
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.
In the paper
“Cui et al. (2016) were the first to introduce the Window Slicing (WS) method as a data augmentation technique.”
What the sources say
- Not checked yetcites Cui et al., 2016
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
“Actually, the difference with the usual convolutions (such as in FCNs) is that a linear shortcut is added to link the output of a residual block to its input thus enabling the flow of the gradient directly through these connections, which makes training a DNN much easier by reducing the vanishing gradient effect (He et al., 2016).”
What the sources say
- Not checked yetcites He et al., 2016
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 problem with an MLP approach is that temporal information is lost and the features learned are no longer time-invariant.”
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
“We made this choice because we believe that the main goal of deep learning approaches is to remove the bias due to manually designed features (Ord´ o¨ nez and Roggen, 2016), thus enabling the network to learn the most discriminant useful features for the classification task.”
What the sources say
- Not checked yetcites Ord´ o¨ nez and Roggen, 2016
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.
In the paper
“This form of pre-processing has also been used by Ratanamahatana and Keogh (2005) to show that the length of a time series is not an issue for TSC problems.”
What the sources say
- Not checked yetcites Ratanamahatana and Keogh (2005)
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.
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
“By using an FC layer before the softmax classifier, the transferability of this network is limited to the first and second convolutional layers.”
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
“The first characteristic of Time-CNN is the use of the mean squared error (MSE) instead of the traditional categorical cross-entropy loss function, which has been used by all the deep learning approaches we have mentioned so far. Hence, instead of a softmax classifier, the final layer is a traditional FC layer with sigmoid as the activation function, which does not guarantee a sum of probabilities”
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
“Unlike GAP, local pooling introduces invariance to small perturbations in the activation map (the result of a convolution) by taking the maximum value in a local pooling window.”
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
“Since we did not train nor test any of the state-of-the-art non deep learning algorithms, it is possible that we allowed much more training time for the described DNNs.”
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.
In the paper
“our findings are in agreement with the deep learning for computer vision literature where deeper neural networks are much more successful than shallower architectures (He et al., 2016).”
What the sources say
- Not checked yetcites He et al., 2016
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
“Interestingly, DiatomSizeReduction is the smallest datasets in the UCR/UEA archive (with 16 training instances), which suggests that ResNet and FCN are easily overfitting this dataset.”
hedgedconditionalSingle 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.
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 step is very important for deep learning models whose architecture depends on the length of the input time series (such as a MLP) and for parallel computation over the GPUs.”
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.
- Rests on an assumption: The MTS archive contains time series of unequal length.
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 to the huge training time, HIVE-COTE's classification time is bounded by a linear scan of the training set due to employing a nearest neighbor classifier, whereas the trivial GPU parallelization of DNNs provides instant classification.”
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.
In the paper
“since increasing the number of convolutional layers will increase the path length viewed by the CNN model (Vaswani et al., 2017), ResNet and FCN managed to outperform other approaches whose filter length is longer (21) such as Encoder.”
What the sources say
- Not checked yetcites Vaswani et al., 2017
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 FC layer reduces the effect of learning time invariant features which explains why MLP, Time-CNN and MCDCNN exhibit very similar performance.”
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
“Therefore, we believe that allowing a much more thorough search of DTW's warping window would lead to a fairer comparison between deep learning approaches and other state-of-the-art TSC algorithms.”
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: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
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
“unlike HIVE-COTE, ResNet's hyperparameters were not tuned for each dataset but rather the same architecture was used for the whole benchmark”
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
“One final observation that is in agreement with the deep learning literature is that in order to achieve high accuracies while training a DNN, a large training set is needed.”
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
“suggesting that these methods are no longer competitive with current state-of-the-art algorithms for TSC.”
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.
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
“suggesting further investigation of these hyperparameters should improve DNNs' accuracy for TSC.”
hedgedconditional- 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.
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 only natural to think that discovering patterns in a one dimensional space (time) should be an easier task for CNNs thus requiring less data to learn from.”
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.
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
“which enables us to take the mean accuracy over the 10 runs in order to reduce the bias due to the weights' initial values.”
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.
In the paper
“for a lazy machine learning algorithm such as NN-DTW, training time is zero when allowing maximum warping whereas it has been shown that judicially setting the warping window Dau et al. (2017) can lead to a significant increase in accuracy.”
What the sources say
- Not checked yetcites Dau et al., 2017
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
“These observations reinforces the need to have an equally large MTS classification archive in order to evaluate hybrid univariate/multivariate time series classifiers.”
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: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
In the paper
“but the fact that ResNet is not significantly different than COTE suggests that more datasets would give a better insight into these performances (Demˇ sar, 2006).”
What the sources say
- Not checked yetcites Demˇ sar, 2006
hedgedconditional- 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.
In the paper
“It is worthwhile noting that cliques formed by the Wilcoxon Signed Rank Test with Holm's alpha correction do not necessary reflect the rank order (Lines et al., 2018).”
What the sources say
- Not checked yetcites Lines et al., 2018 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.
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 Time-CNN and MCDCNN are the only architectures originally proposed for MTS data”
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
“We believe that the success of ResNet is highly due to its deep flexible architecture.”
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: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
In the paper
“especially since normalization is known to have a huge effect on DNNs' learning capabilities (Zhang et al., 2017).”
What the sources say
- Not checked yetcites Zhang et al., 2017
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
“which probably explains why FCN is less accurate than ResNet on this dataset.”
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.
The quote only partly states this claim.
- Rests on an assumption: CAM visualizations faithfully indicate which time series regions contribute to class identification.
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 fact, most DNNs architectures, with the categorical cross-entropy as their cost function, employ mainly the same classifier: softmax which is basically designed for multi-class classification.”
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.
In the paper
“we stress that the shapelet transformation classifier (Hills et al., 2014) is an ensemble approach, which makes unclear how the shapelets affect the decision taken by the individual classifiers whereas for an end-to-end deep learning model we can directly explain the classification by using the Class Activation Map.”
What the sources say
- Not checked yetcites Hills et al., 2014 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
“On the other hand, the part where the actor raises his/her arm seems to be also a discriminative part of the data which suggests that the deep learning algorithms are identifying more 'shapelets'.”
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. 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
“Finally, another interesting observation is the ability of CNNs to localize a given discriminative shape regardless where it appears in the time series, which is evidence for CNNs' capability of learning time-invariant warped features.”
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
“We should note that this observation cannot confirm which classifier extracted the most discriminative subsequences especially because all algorithms achieved similar accuracy on GunPoint dataset.”
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 relies on this without support.
The UCR/UEA archive datasets are representative of TSC problems.
“we tested the approaches on 12 Multivariate Time Series (MTS) datasets (Baydogan, 2015).”
The paper relies on this without support.
The 97 time series datasets used are sufficient to generalize about DNN performance for TSC.
“By training 8,730 deep learning models on 97 time series datasets, we propose the most exhaustive study of DNNs for TSC to date.”
The paper relies on this without support.
The deep learning implementations in the framework faithfully represent the original methods.
“We also provide an open source deep learning framework to the TSC community where we implemented each of the compared approaches and evaluated them on a univariate TSC benchmark (the UCR/UEA archive) and 12 multivariate time series datasets.”
The paper relies on this without support.
The MTS archive contains time series of unequal length.
“In order to solve the problem of unequal length time series in the MTS archive”
The paper relies on this without support.
CAM visualizations faithfully indicate which time series regions contribute to class identification.
“Highlighting with the Class Activation Map the contribution of each time series region for the three classes in Meat when using the FCN and ResNet classifiers.”
The paper relies on this without support.
MDS plots with Euclidean distance on GAP vectors preserve class separability sufficiently to interpret model performance.
“Figure 15 shows three MDS plots for the GunPoint dataset using: (1) the raw input time series (Figure 15a); (2) the learned latent features from the GAP layer for FCN (Figure 15b); and (3)”
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
“These approaches, categorized as temporal representation[105,106], divide the entire framework into two stages: “temporal representation” and “temporal prediction.””
From Advancements in the estimation of the state of charge of lithium-ion battery: a comprehensive review of traditional and deep learning approaches (2025) · cites this paper as [105]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“These include the high computational cost incurred by running multiple instances [18,19]”
From TVGeAN: Tensor Visibility Graph-Enhanced Attention Network for Versatile Multivariant Time Series Learning Tasks (2024) · cites this paper as [19]
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
§ 4.1.1 Univariate archive · p. 21
“| WalkVsRun | 128-1918 | 1919 | 2 | 62 | 28 | 16 |”
§ 4.1.2 Multivariate archive · p. 21
“thus each time series will have a length equal to the longest time series' length.”
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
§ 3.2.1 Multi Layer Perceptron · p. 14
“MLP does not have any layer whose number of parameters is invariant across time series of different lengths”
§ 3.2.1 Multi Layer Perceptron · p. 14
“All three hidden FC layers are composed of 500 neurons with ReLU as the activation function.”
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 12×
“First of all, our findings are in agreement with the deep learning for computer vision literature where deeper 1 The implementations are available on https://github.com/hfawaz/dl-4-tsc neural networks are much more successful than shallower architectures (He et al., 2016).”
BackgroundResultInfluentialVaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017) · arXiv · doi:10.48550/arxiv.1706.03762 · Cited by 120,404 · cited 4×
“The MLP, which is the most traditional form of DNNs, was proposed in Wang et al. (2017b) as a baseline architecture for TSC.”
BackgroundMethodInfluentialImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155 · cited 7×
“Since AlexNet (Krizhevsky et al. 2012) won the ImageNet competition in 2012, deep CNNs have seen a lot of successful applications in many different domains (LeCun et al.”
BackgroundInfluentialAdam: A Method for Stochastic Optimization
Kingma, D. P., & Ba, J. (2014) · arXiv · doi:10.48550/arxiv.1412.6980 · Cited by 98,946 · cited 1×
“All models were optimized using a variant of Stochastic Gradient Descent (SGD) such as Adam (Kingma and Ba, 2015) and AdaDelta (Zeiler, 2012).”
MethodLeCun, Y., Bengio, Y., & Hinton, G. (2015) · Nature · doi:10.1038/nature14539 · Cited by 84,126
Gradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805
Going deeper with convolutions
Szegedy, C., Liu, W., Jia, Y., Sermanet, P., Reed, S., Anguelov, D., Erhan, D., Vanhoucke, V., & Rabinovich, A. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298594 · Cited by 48,562
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380
(2014) · Cited by 33,419
Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
Ioffe, S., & Szegedy, C. (2015) · arXiv · doi:10.48550/arxiv.1502.03167 · Cited by 27,818
Research works printed but not linked: 2
- Che Z, He X, Xu K, Liu Y (2017b) DECADE: a deep metric learning model for multivariate time series. In: KDD workshop on mining and learning from time series
- Ismail Fawaz H, Forestier G, Weber J, Idoumghar L, Muller PA (2018c) Transfer learning for time series classification. In: IEEE international conference on big data, pp 1367–1376 · DOI · Paper page
Cited by
Sorted by citations
Items 601–700 of 1,007
Live Road Condition Assessment with Internal Vehicle Sensors
Levenberg, E., Skar, A., Pour, S. M., Kindler, E., Pettinari, M., Bajic, M., Alstrøm, T. S., & Schlotz, U. (2021) · Transportation Research Record Journal of the Transportation Research Board · doi:10.1177/03611981211016852 · Cited by 24
“Different machine learning models are evaluated in separate settings, with primary attention given to supervised learning of time-series data using deep neural networks (34).”
BackgroundNew universal sustainability metrics to assess edge intelligence
Lenherr, N., Pawlitzek, R., & Michel, B. (2021) · Sustainable Computing Informatics and Systems · doi:10.1016/j.suscom.2021.100580 · Cited by 24
Attention-Aware Dynamic Self-Aggregation Network for Satellite Image Time Series Classification
Zhang, W., Du, P., Fu, P., Zhang, P., Tang, P., Zheng, H., Meng, Y., & Li, E. (2021) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2021.3120914 · Cited by 26
Thoppil, N. M., Vasu, V., & Rao, C. S. P. (2021) · Journal of Computing and Information Science in Engineering · doi:10.1115/1.4052838 · Cited by 25
Role of Data Augmentation Strategies in Knowledge Distillation for Wearable Sensor Data
Jeon, E. S., Som, A., Shukla, A., Hasanaj, K., Buman, M. P., & Turaga, P. K. (2021) · IEEE Internet of Things Journal · doi:10.1109/jiot.2021.3139038 · Cited by 25
“In general, stacking more layers or increasing the number of learnable parameters causes deep networks to exhibit improved performance [2]–[4], [8]–[10].”
Meteier, Q., de Salis, E., Capallera, M., Widmer, M., Angelini, L., Khaled, O. A., Sonderegger, A., & Mugellini, E. (2022) · Frontiers in Computer Science · doi:10.3389/fcomp.2021.775282 · Cited by 24
Financial markets’ deterministic aspects modeled by a low-dimensional equation
Orlando, G., Bufalo, M., & Stoop, R. (2022) · Scientific Reports · doi:10.1038/s41598-022-05765-z · Cited by 23
Ott, F., Rügamer, D., Heublein, L., Bischl, B., & Mutschler, C. (2022) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv51458.2022.00131 · Cited by 24
“An overview of MTSC meth-ods can be found in [22].”
BackgroundHssayni, E. H., Joudar, N., & Ettaouil, M. (2022) · Computational Intelligence · doi:10.1111/coin.12556 · Cited by 24
“12 In recent years, as a result of the extraordinary development of more machine learning fields, deep neural networks becoming more and more important in the classification of time series.”
BackgroundPredicting readmission to the cardiovascular intensive care unit using recurrent neural networks
Kessler, S., Schroeder, D., Korlakov, S., Hettlich, V. H., Kalkhoff, S., Moazemi, S., Lichtenberg, A., Schmid, F., & Aubin, H. (2023) · Digital Health · doi:10.1177/20552076221149529 · Cited by 24
“Another possible way to improve performance might be utilizing state-of-the-art models such as residual networks(32) which have shown to achieve good results for multivariate time-series classification and prediction.(33) This study has further limitations.”
MethodPark, J., Lim, W.-S., Kim, D., & Lee, J. (2023) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2023.106543 · Cited by 23
An ensemble machine learning approach for classification tasks using feature generation
Feng, W., Gou, J., Fan, Z., & Chen, X. '. (2023) · Connection Science · doi:10.1080/09540091.2023.2231168 · Cited by 24
“Thus, we used the critical difference diagram with Wilcoxon-Holm method (Ismail Fawaz et al., 2019) to show the performance effect of these algorithms in terms of Accuracy and F1-score while the statistical results ( p -values) based on the Wilcoxon-Holm method to detect pairwise significance are…”
MethodClassification of Human Motion Data Based on Inertial Measurement Units in Sports: A Scoping Review
Kranzinger, C., Bernhart, S., Kremser, W., Venek, V., Rieser, H., Mayr, S., & Kranzinger, S. (2023) · Applied Sciences · doi:10.3390/app13158684 · Cited by 26
Viana, J., Farkhari, H., Sebastião, P., Campos, L. M., Koutlia, K., Bojović, B., Lagén, S., & Dinis, R. (2023) · IEEE Transactions on Vehicular Technology · doi:10.1109/tvt.2023.3302814 · Cited by 24
“In DL, Deep Neural Networks (DNNs), such as Convolutional Neural Networks (CNNs), can define trends and seasonality in time series data [26], [27], [28].”
BackgroundPeláez-Rodríguez, C., Pérez‐Aracil, J., Fister, D., Torres-López, R., & Salcedo‐Sanz, S. (2023) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.122264 · Cited by 27
A Deep Convolutional Neural Network for Time Series Classification with Intermediate Targets
Taherkhani, A., Cosma, G., & McGinnity, T. M. (2023) · SN Computer Science · doi:10.1007/s42979-023-02159-4 · Cited by 30
“…expensive than training a CNN. Additionally, recent research has shown that certain convolutional architectures for different applications, such as audio synthesis, machine translation, and skeleton-based action prediction, can reach state-of-the-art accuracies [15, 17, 19, 21, 48, 58].”
BackgroundMethodInfluentialBen Chaabane, N., Conze, P.-H., Lempereur, M., Quellec, G., Rémy-Néris, O., Brochard, S., Cochener, B., & Lamard, M. (2023) · Scientific Reports · doi:10.1038/s41598-023-49883-8 · Cited by 24
“The obtained results were compared to five state-of-the-art architectures 10 , including Fully Convolutional neural Network (FCN), Residual Network (ResNet), Encoder, Time Le-Net (t-LeNet), and Transformer.”
ResultData-driven blood glucose level prediction in type 1 diabetes: a comprehensive comparative analysis
Nemat, H., Khadem, H., Elliott, J., & Benaissa, M. (2024) · Scientific Reports · doi:10.1038/s41598-024-70277-x · Cited by 29
Yang, F., Abedin, M. Z., Hájek, P., & Qiao, Y. (2025) · IEEE Transactions on Engineering Management · doi:10.1109/tem.2025.3547730 · Cited by 23
Zhang, Y.-H., & Zhao, H. (2020) · Canadian journal of remote sensing · doi:10.1080/07038992.2020.1740083 · Cited by 19
“2012) and (2) unsupervised classification based on the time series images (Fawaz et al. 2019; Paris et al. 2019).”
BackgroundMethodShapelet-transformed Multi-channel EEG Channel Selection
Dai, C., Pi, D., & Becker, S. I. (2020) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/3397850 · Cited by 24
“Honestly, there are many classifiers for EEG time-series classification, including Deep Neural Networks (DNNs)-based classifier [13], COTE [5], HIVE-COTE [34], St-TSC [33], RPCD [46], and SAX-SEQL [39], but we just apply SVM classifier in the article, since (1) SVM is the most widely used and…”
MethodA Time Series Data Augmentation Method Based on Dynamic Time Warping
Yang, X., Zhang, Z., Cui, X., & Cui, R. (2021) · International Conference on Computer Communication and Artificial Intelligence (CCAI) · doi:10.1109/ccai50917.2021.9447507 · Cited by 23
ForecastQA: A Question Answering Challenge for Event Forecasting with Temporal Text Data
Jin, W., Khanna, R., Kim, S., Lee, D., Morstatter, F., Galstyan, A., & Ren, X. (2021) · Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2021.acl-long.357 · Cited by 28
Detection of Unilateral Arm Paresis after Stroke by Wearable Accelerometers and Machine Learning
Wassélius, J., Finn, E. L., Persson, E., Ericson, P., Brogårdh, C. B., Lindgren, A. G., Ullberg, T., & Åström, K. (2021) · Sensors · doi:10.3390/s21237784 · Cited by 25
“The TensorFlow implementation of FCN available in the companion repository to [38] was used in this study (https://github.”
BackgroundMethodVIMSNet: an effective network for visually induced motion sickness detection
Liu, R., Cui, S., Zhao, Y., Chen, X., Lin, Y., & Hwang, A. D. (2022) · Signal Image and Video Processing · doi:10.1007/s11760-022-02164-9 · Cited by 24
“FCN and ResNet are both CNN models which have been widely used as strong baselines for EEG applications [29,30].”
MethodDetecting Smartwatch-Based Behavior Change in Response to a Multi-Domain Brain Health Intervention
Cook, D. J., Strickland, M., & Schmitter‐Edgecombe, M. (2022) · ACM Transactions on Computing for Healthcare · doi:10.1145/3508020 · Cited by 23
“, a change from sedentary to active behavior), but these algorithms must be supplied with a sufficient amount of labeled training data [35].”
BackgroundBier, A., Jastrzębska, A., & Olszewski, P. (2022) · IEEE Access · doi:10.1109/access.2022.3203523 · Cited by 23
Vehicle theft detection by generative adversarial networks on driving behavior
Tseng, P.-Y., Lin, P., & Kristianto, E. (2022) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2022.105571 · Cited by 23
A fuzzy-based cascade ensemble model for improving extreme wind speeds prediction
Peláez-Rodríguez, C., Pérez‐Aracil, J., Prieto-Godino, L., Ghimire, S., Deo, R. C., & Salcedo‐Sanz, S. (2023) · Journal of Wind Engineering and Industrial Aerodynamics · doi:10.1016/j.jweia.2023.105507 · Cited by 23
Eddy, E., Campbell, E., Bateman, S., & Scheme, E. (2024) · Journal of Neural Engineering · doi:10.1088/1741-2552/ad4915 · Cited by 23
Effect of Data Representation for Time Series Classification—A Comparative Study and a New Proposal
Nakano, K., & Chakraborty, B. (2019) · Machine Learning and Knowledge Extraction · doi:10.3390/make1040062 · Cited by 25
“CNN and ResNet are known to be the most successful and effective among deep neural networks (DNN) so far according to Reference [20].”
BackgroundCombining contextual neural networks for time series classification
Kamara, A. F., Chen, E., Liu, Q., & Pan, Z. (2019) · Neurocomputing · doi:10.1016/j.neucom.2019.10.113 · Cited by 23
Exploring Time Series Imaging for Load Disaggregation
Bousbiat, H., Klemenjak, C., & Elmenreich, W. (2020) · ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys) · doi:10.1145/3408308.3427975 · Cited by 23
“Inspired by the success of time-series imaging in related research elds [7, 10], we suspect that looking beyond one-dimensional NILM could bear the potential to close this performance gap.”
BackgroundMulti-modal lifelog data fusion for improved human activity recognition: A hybrid approach
Oh, Y., & Kim, S. (2024) · Information Fusion · doi:10.1016/j.inffus.2024.102464 · Cited by 22
A novel gradient boosting approach for imbalanced regression
Zhang, W., Shi, P., Jia, P., & Zhou, X. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.128091 · Cited by 22
Low-dimensional Convolutional Neural Network for Solar Flares GOES Time-series Classification
Landa, V., & Reuveni, Y. (2022) · The Astrophysical Journal Supplement Series · doi:10.3847/1538-4365/ac37bc · Cited by 27
Dabou, R. T., Kamwa, I., Chung, C. Y., & Mugombozi, C. F. (2021) · IEEE Access · doi:10.1109/access.2021.3099459 · Cited by 22
Complex Valued Deep Neural Networks for Nonlinear System Modeling
Lopez-Pacheco, M. A., & Yu, W. (2021) · Neural Processing Letters · doi:10.1007/s11063-021-10644-1 · Cited by 22
LSTM-MFCN: A time series classifier based on multi-scale spatial–temporal features
Zhao, L., Mo, C., Ma, J., Chen, Z., & Yao, C. (2021) · Computer Communications · doi:10.1016/j.comcom.2021.10.036 · Cited by 22
Personalized human activity recognition using deep learning and edge-cloud architecture
Alawneh, L. M., Al‐Ayyoub, M., Al-Sharif, Z. A., & Shatnawi, A. S. (2022) · Journal of Ambient Intelligence and Humanized Computing · doi:10.1007/s12652-022-03752-w · Cited by 21
Kumar, A., & Gorai, A. K. (2022) · Geocarto International · doi:10.1080/10106049.2022.2057595 · Cited by 24
Tnani, M.-A., Subarnaduti, P., & Diepold, K. J. (2022) · Sensors · doi:10.3390/s22134813 · Cited by 22
“Unlike image processing, research in machine monitoring has mostly overlooked the advantages of deep neural networks due to their hard interpretability and their nonacceptance in industry [9].”
BackgroundBoniol, P., Paparrizos, J., Kang, Y., Palpanas, T., Tsay, R. S., Elmore, A. J., & Franklin, M. J. M. (2022) · Proceedings of the VLDB Endowment · doi:10.14778/3554821.3554879 · Cited by 22
TSEvo: Evolutionary Counterfactual Explanations for Time Series Classification
Höllig, J., Kulbach, C., & Thoma, S. (2022) · International Conference on Machine Learning and Applications · doi:10.1109/icmla55696.2022.00013 · Cited by 22
“Both types of model-specific methods lack the application to a significant fraction of the widespread machine learning techniques used for time series classification (e.g., [3], [10]).”
BackgroundMethodA novel method for aero-engine time-series forecasting based on multi-resolution transformer
Jin, H.-J., Zhao, Y.-P., & Pan, M.-N. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.124597 · Cited by 23
Neural network developments: A detailed survey from static to dynamic models
Verma, P. R., Singh, N. P., Pantola, D., & Cheng, X. (2024) · Computers & Electrical Engineering · doi:10.1016/j.compeleceng.2024.109710 · Cited by 26
Estimation of neural network model parameters from local field potentials (LFPs)
Skaar, J.-E. W., Stasik, A., Hagen, E., Ness, T. V., & Einevoll, G. T. (2020) · PLoS Computational Biology · doi:10.1371/journal.pcbi.1007725 · Cited by 20
“Other machine learning techniques, see [53] for a recent review, could likely have done equally well, or even better.”
BackgroundVillalobos, K., Suykens, J. A. K., & Illarramendi, A. (2020) · Journal of Intelligent Manufacturing · doi:10.1007/s10845-020-01614-w · Cited by 23
“Furthermore, if due to special requirements of the application scenario more precision is required for a particular sensor, a specific forecaster could be built to predict only the future measurements of that sensor in a more accurately way.”
BackgroundMethodInfluentialTrustworthy Building Fire Detection Framework With Simulation-Based Learning
Kim, Y.-J., Kim, H., Lee, S.-G., & Kim, W.-T. (2021) · IEEE Access · doi:10.1109/access.2021.3071552 · Cited by 24
“The FCN and ResNet models have convolutional architectures that show the highest performance on various multivariate time-series datasets [40].”
BackgroundMethodInfluentialSoleimani, R., & Lobaton, E. J. (2022) · Information · doi:10.3390/info13070326 · Cited by 21
Physiology-Driven Empathic Large Language Models (EmLLMs) for Mental Health Support
Dongre, P. (2024) · CHI Conference Extended Abstracts on Human Factors in Computing Systems (CHI EA) · doi:10.1145/3613905.3651132 · Cited by 21
Liebers, J., Laskowski, P., Rademaker, F., Sabel, L., Hoppen, J., Gruenefeld, U., & Schneegaß, S. (2024) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3613904.3642471 · Cited by 21
“We selected these models since they are founded in the review on time-series classification by Fawaz et al., and we also utilized their implementation with default hyperparameters 13 [21].”
MethodMgformer: Multi-group transformer for multivariate time series classification
Wen, J., Zhang, N., Lu, X., Hu, Z., & Huang, H. (2024) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2024.108633 · Cited by 25
Extraction of Features for Time Series Classification Using Noise Injection
Kim, G.-I., & Chung, K. (2024) · Sensors · doi:10.3390/s24196402 · Cited by 24
“Existing time series classification methods include traditional statistical techniques [4], machine learning algorithms [5], and deep learning-based models [6], but they have several limitations due to data imbalance, noise, and limited data quantities [7].”
BackgroundVessel Trajectory Data Mining: A Review
Troupiotis-Kapeliaris, A., Kastrisios, C., & Zissis, D. (2025) · IEEE Access · doi:10.1109/access.2025.3525952 · Cited by 21
ALGAN: Time Series Anomaly Detection with Adjusted-LSTM GAN
Bashar, M. A., & Nayak, R. (2025) · International Journal of Data Science and Analytics · doi:10.1007/s41060-025-00810-2 · Cited by 33
“The critical difference has been proven to be an effective tool for ranking the performance of models, particularly when the number of datasets for evaluation is high [45].”
Posture detection using Deep Learning for Time Series Data
Gupta, R., Saini, D., & Mishra, S. K. (2020) · International Conference on Smart Systems and Inventive Technology (ICSSIT) · doi:10.1109/icssit48917.2020.9214223 · Cited by 22
“Deep learning has also been used on time-series data to classify human activity [10-14].”
MethodGeneralised gravitational wave burst generation with generative adversarial networks
Mcginn, J., Messenger, C., Williams, M. J., & Heng, I. S. (2021) · Classical and Quantum Gravity · doi:10.1088/1361-6382/ac09cc · Cited by 21
“Although most work with CNNs involves image-based data, they can be applied to other spatially adjacent data types such as time-series [49] and text items [50].”
BackgroundSeong, N., & Nam, K. (2021) · Knowledge-Based Systems · doi:10.1016/j.knosys.2021.107608 · Cited by 21
TSFuse: automated feature construction for multiple time series data
Brabandere, A. D., Beéck, T. O. D., Hendrickx, K., Meert, W., & Davis, J. J. (2022) · Machine Learning · doi:10.1007/s10994-021-06096-2 · Cited by 23
Autoencoder Based Anomaly Detection and Explained Fault Localization in Industrial Cooling Systems
Holly, S., Heel, R., Katic, D., Schoeffl, L., Stiftinger, A., Holzner, P., Kaufmann, T., Haslhofer, B., Schall, D., Heitzinger, C., et al. (2022) · PHM Society European Conference · doi:10.36001/phme.2022.v7i1.3349 · Cited by 21
Hybrid Time-Series Forecasting Models for Traffic Flow Prediction
Rajalakshmi, V., & Ganesh Vaidyanathan, S. (2022) · PROMET - Traffic&Transportation · doi:10.7307/ptt.v34i4.3998 · Cited by 21
“Different machine learning and deep learning models are utilised in many researches to reduce the forecasting errors and maximise the precision value [10–12].”
AutoTransformer: Automatic Transformer Architecture Design for Time Series Classification
Ren, Y., Li, L., Yang, X., & Zhou, J. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-05933-9_12 · Cited by 21
Broad fuzzy cognitive map systems for time series classification
Wu, K., Yuan, K., Teng, Y., Liu, J., & Jiao, L. (2022) · Applied Soft Computing · doi:10.1016/j.asoc.2022.109458 · Cited by 21
Near-Infrared Spectroscopy for Bladder Monitoring: A Machine Learning Approach
Fechner, P., König, F., Kratsch, W., Lockl, J., & Röglinger, M. (2022) · ACM Transactions on Management Information Systems · doi:10.1145/3563779 · Cited by 24
TransDBC: Transformer for Multivariate Time-Series based Driver Behavior Classification
Vyas, J., Bhardwaj, N., Bhumika, B., & Das, D. (2022) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn55064.2022.9892146 · Cited by 21
Kim, J., Kim, H., Lee, J., Lee, J., Yoon, J., & Ko, S. (2022) · IEEE Access · doi:10.1109/access.2022.3205112 · Cited by 21
“Inspired by several studies that applied 90 Convolutional Neural Networks (CNN) to sequential data 91 [24], [25], [26], we suggest a deep neural architecture called 92 the FatigueNet that consists of deep convolutional layers 93 followed by three Gated Recurrent Units (GRU) which takes 94 raw time-series spatio-tempor…”
MethodBenchmarking deep learning approaches for all-vs-all conjunction screening
Stevenson, E., Rodríguez-Fernández, V., Urrutxua, H., & Camacho, D. (2023) · Advances in Space Research · doi:10.1016/j.asr.2023.01.036 · Cited by 21
Buss, E., Aust, T., Wahby, M., Rabbel, T.-L., Kernbach, S., & Hamann, H. (2023) · Bioinspiration & Biomimetics · doi:10.1088/1748-3190/acbad2 · Cited by 20
“We trained ten different artificial neural network (ANN) classifiers based on the machine learning framework by Fawaz et al [15].”
BackgroundMethodInfluentialThe use of deep learning for smartphone-based human activity recognition
Stampfler, T., Elgendi, M., Fletcher, R., & Menon, C. (2023) · Frontiers in Public Health · doi:10.3389/fpubh.2023.1086671 · Cited by 21
“We started from the base Resnet architecture of Ismail Fawaz et al. (13), which is available at the following GitHub repository: dl-4-tsc.”
BackgroundMethodInfluentialDCFF-MTAD: A Multivariate Time-Series Anomaly Detection Model Based on Dual-Channel Feature Fusion
Xu, Z., Yang, Y., Gao, X., & Hu, M. (2023) · Sensors · doi:10.3390/s23083910 · Cited by 24
Knap, P., Lalik, K., & Bałazy, P. (2023) · Sensors · doi:10.3390/s23094295 · Cited by 23
“The models perform inference based on the three-channel input, using operations for every channel [38].”
BackgroundMethodSpectral Cross-Domain Neural Network With Soft-Adaptive Threshold Spectral Enhancement
Liu, C., Cheng, S. L., Ding, W., & Arcucci, R. (2023) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2023.3332217 · Cited by 28
“O SATSE i Output of the SATSE block.”
Ashfaq, N., Khan, M. H., & Nisar, M. A. (2024) · Information · doi:10.3390/info15060343 · Cited by 25
Artificial neural network subgrid models of 2D compressible magnetohydrodynamic turbulence
Rosofsky, S., & Huerta, E. A. (2020) · Physical review. D/Physical review. D. · doi:10.1103/physrevd.101.084024 · Cited by 28
Nonlinear Noise Cleaning in Gravitational-Wave Detectors With Convolutional Neural Networks
Yu, H., & Adhikari, R. X. (2022) · Frontiers in Artificial Intelligence · doi:10.3389/frai.2022.811563 · Cited by 26
Speech Driven Gaze in a Face-to-Face Interaction
Aydın, Ü. A., Kalkan, S., & Acartürk, C. (2021) · Frontiers in Neurorobotics · doi:10.3389/fnbot.2021.598895 · Cited by 21
“Although CNN models are mostly used for image processing, they can be used in the same manner for time series (Fawaz et al., 2019).”
MethodWang, D., Hu, Q., & Yang, C. (2022) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2022.105654 · Cited by 20
Widiasari, I. R., & Efendi, R. (2024) · Informatics · doi:10.3390/informatics11040073 · Cited by 20
“Some algorithms that are often used for prediction are long short-term memory (LSTM) and gated recurrent unit (GRU), which have been confirmed impactful in time series data analysis because of their ability to handle temporal dependencies [11–13].”
Data-driven surrogate model for wind turbine damage equivalent load
Haghi, R., & Crawford, C. A. (2024) · Wind energy science · doi:10.5194/wes-9-2039-2024 · Cited by 23
“Research has shown that the TCN is better than the recurrent neural network (RNN) and LSTM in terms of performance, implementation, flexibility and versatility (Fawaz et al., 2019; Bai et al., 2018).”
BackgroundA Comprehensive Survey on Multi-View Classification: Methods, Applications, and Challenges
Berahmand, K., Daneshfar, F., Rahmaninia, M., Haghighat, M., & Jalili, M. (2025) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/3767728 · Cited by 27
“MVC is a technique in computer vision and 3D shape analysis that improves object recognition by capturing multiple viewpoints of a point cloud, offering a more comprehensive understanding of its shape [98].”
Spiller, M., Liu, Y., Hossain, M. Z., Gedeon, T., Geissler, J., & Nürnberger, A. (2021) · ACM Transactions on Interactive Intelligent Systems · doi:10.1145/3446638 · Cited by 20
fNIRS Signal Classification Based on Deep Learning in Rock-Paper-Scissors Imagery Task
Ma, T., Chen, W., Li, X. Y., Xia, Y., Zhu, X., & He, S. (2021) · Applied Sciences · doi:10.3390/app11114922 · Cited by 22
Arc Fault Detection in DC Microgrid Using Deep Neural Network
Patil, D., Bindu, S., & SushilThale (2021) · Biennial International Conference on Nascent Technologies in Engineering (ICNTE) · doi:10.1109/icnte51185.2021.9487585 · Cited by 19
Potential of Rule-Based Methods and Deep Learning Architectures for ECG Diagnostics
Bortolan, G., Christov, I. I., & Simova, I. I. (2021) · Diagnostics · doi:10.3390/diagnostics11091678 · Cited by 20
“There has been much focus on research employing the use of deep learning with medical images [4], time series classification [5], and object detection [6].”
BackgroundMultivariate temporal data analysis ‐ a review
Moskovitch, R. (2021) · Wiley Interdisciplinary Reviews Data Mining and Knowledge Discovery · doi:10.1002/widm.1430 · Cited by 21
Learning Time Series Counterfactuals via Latent Space Representations
Wang, Z., Samsten, I., Mochaourab, R., & Papapetrou, P. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-88942-5_29 · Cited by 21
Evaluation of Deep Learning Methods in a Dual Prediction Scheme to Reduce Transmission Data in a WSN
Morales, C. R., de Sousa, F. R., Brusamarello, V. J., & Fernandes, N. C. (2021) · Sensors · doi:10.3390/s21217375 · Cited by 21
“One-Dimensional Convolutional Neural Networks (1D-CNNs), Recurrent Neural Networks (RNNs), Long Short-Term Memory Neural Networks (LSTMs), Gated Recurrent Unit (GRU) and combinations of these architectures have been used in many classification and forecasting problems with sequential data [17,18].”
MethodCloete, J. B., Stander, T., & Wilke, D. N. (2022) · IEEE Access · doi:10.1109/access.2022.3149324 · Cited by 20
Mantach, S., Gill, P., Oliver, D. R., Ashraf, A. B., & Kordi, B. (2022) · Neural Computing and Applications · doi:10.1007/s00521-022-07066-y · Cited by 20
Adhikari, A., Naetiladdanon, S., & Sangswang, A. (2022) · Applied Sciences · doi:10.3390/app12136333 · Cited by 23
Deep-Learning-Based Multivariate Time-Series Classification for Indoor/Outdoor Detection
Bakirtzis, S., Qiu, K., Wassell, I. J., Fiore, M., & Zhang, J. (2022) · IEEE Internet of Things Journal · doi:10.1109/jiot.2022.3190555 · Cited by 20
“Assuming k classes, each time-series sample X i is associated to a certain class to form an input-target pair, and in the output, the DL model predicts the probability of X i belonging to each class [34].”
BackgroundTowards Backdoor Attack on Deep Learning based Time Series Classification
Ding, D., Zhang, M., Huang, Y., Pan, X., Feng, F., Jiang, E., & Yang, M. (2022) · IEEE International Conference on Data Engineering (ICDE) · doi:10.1109/icde53745.2022.00100 · Cited by 22
Borovits, N., Kumara, I., Nucci, D. D., Krishnan, P., Palma, S. D., Palomba, F., Tamburri, D. A., & van den Heuvel, W. (2022) · Empirical Software Engineering · doi:10.1007/s10664-022-10215-5 · Cited by 22
“To perform this statistical analysis and draw CD diagrams, we relied on the implementation provided by Ismail Fawaz et al. (2019).”
MethodLearning the micro-environment from rich trajectories in the context of mobile crowd sensing
Hafyani, H. E., Abboud, M., Zuo, J., Zeitouni, K., Taher, Y., Chaix, B., & Wang, L. (2022) · GeoInformatica · doi:10.1007/s10707-022-00471-4 · Cited by 20
“There is a wide range of time series classification approaches that can be classified into four categories: distance-based methods [4], feature-based methods [35], ensemble methods [21], and deep learning models [9, 19, 46].”
Explainable machine learning for breakdown prediction in high gradient rf cavities
Obermair, C., Cartier-Michaud, T., Apollonio, A., Millar, W. L., Felsberger, L., Fischl, L., Bovbjerg, H. S., Wollmann, D., Wuensch, W., Lasheras, N. C., et al. (2022) · Physical Review Accelerators and Beams · doi:10.1103/physrevaccelbeams.25.104601 · Cited by 23
Time-frequency based multi-task learning for semi-supervised time series classification
Wei, C., Wang, Z., Yuan, J., Li, C., & Chen, S. (2022) · Information Sciences · doi:10.1016/j.ins.2022.11.040 · Cited by 20
Xu, Y., Lian, G., Zhou, H., Hou, Y., Zhang, H., Zhang, L., Yan, R., & Chen, X. (2022) · Composites Science and Technology · doi:10.1016/j.compscitech.2022.109853 · Cited by 21
Yang, W., Liu, W., & Gao, Q. (2022) · Mathematical Biosciences & Engineering · doi:10.3934/mbe.2023046 · Cited by 20
Naku, W., Nambisan, A. K., Roman, M., Zhu, C., Gerald, R. E., & Huang, J. (2023) · ACS Omega · doi:10.1021/acsomega.2c05451 · Cited by 22
Metadata
- Authors
- Ismail Fawaz, H., Forestier, G., Weber, J., Idoumghar, L., & Muller, P.-A.
- Affiliations
- Université de Haute-Alsace (FR)
- Venue
- Data Mining and Knowledge Discovery
- Published
- 2019-03-02
- Volume
- 33
- Issue
- 4
- Pages
- 917-963
- ISSN
- 1384-5810 (ISSN-L)
- 1573-756X
- OpenAlex
- W2892035503
- Cited by
- 3,342
- References
- 149
- Citations by year
- 2026: 385 · 2025: 493 · 2024: 507 · 2023: 517 · 2022: 527 · 2021: 494 · 2020: 328 · 2019: 85
- FWCI
- 227.10
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- Data Mining and Knowledge Discovery (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- HAL Portal Artxiker (Hindustan Aeronautics Limited (India)) (submittedVersion, other-oa)
- HAL (submittedVersion, other-oa)
- arXiv (Cornell University)
- Funders
- Agence Nationale de la Recherche (ANR-17-CE23-0015)
- Topics
- Physical Sciences › Computer Science › Signal Processing › Time Series Analysis and Forecasting
- Physical Sciences › Computer Science › Signal Processing › Music and Audio Processing
- Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Applications
- Keywords
- time series classification, deep neural network, deep learning architectures, convolutional neural networks, ResNet, multivariate time series, univariate time series
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 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-09 (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
