The DOI matches its Crossref record; 38 of the 38 references it cites were found in a registry or an index; none of the 13 sources its claims cite is checked for support yet.
Tree-CNN: A hierarchical Deep Convolutional Neural Network for incremental learning
Open accessAbstract
Over the past decade, Deep Convolutional Neural Networks (DCNNs) have shown remarkable performance in most computer vision tasks. These tasks traditionally use a fixed dataset, and the model, once trained, is deployed as is. Adding new information to such a model presents a challenge due to complex training issues, such as "catastrophic forgetting", and sensitivity to hyper-parameter tuning. However, in this modern world, data is constantly evolving, and our deep learning models are required to adapt to these changes. In this paper, we propose an adaptive hierarchical network structure composed of DCNNs that can grow and learn as new data becomes available. The network grows in a tree-like fashion to accommodate new classes of data, while preserving the ability to distinguish the previously trained classes. The network organizes the incrementally available data into feature-driven super-classes and improves upon existing hierarchical CNN models by adding the capability of self-growth. The proposed hierarchical model, when compared against fine-tuning a deep network, achieves significant reduction of training effort, while maintaining competitive accuracy on CIFAR-10 and CIFAR-100.
Overview
AI snapshot
Written from the paper’s full text.
TL;DR
Tree-CNN is a hierarchical DCNN that grows tree-like as new classes arrive, assigning new classes to existing, merged, or new child nodes based on softmax likelihood and retraining only affected nodes. On CIFAR-10 and CIFAR-100 it reduces training effort versus fine-tuning a baseline network while maintaining competitive accuracy.
Key findings
- On CIFAR-10, after adding 4 new classes, Tree-CNN reached 86.24% test accuracy with normalized training effort 0.60, versus baseline B:V at 90.51% accuracy and effort 1.00; Tree-CNN used 40% less effort than B:V and 30% less than B:II, while being 3.76% below B:V.
- On CIFAR-100 over 9 incremental stages, Tree-CNN-5/10/20 achieved final test accuracies of 61.57%, 60.46%, and 59.99% (average 69.85%, 69.53%, 68.49%), compared with B:V at 63.05% final / 72.23% average, iCaRL at 49.11% / 64.10%, and LwF at 25.00% / 44.49%. The paper reports Tree-CNNs yield 10% higher accuracy than iCaRL and over 50% higher than LwF.
- Tree-CNNs required lower training effort than fine-tuning cases B:II–B:V; B:I had the lowest effort but the worst accuracy degradation. Tree-CNN-5 grew to 3.4× the size of Network B, Tree-CNN-10 to 2.2×, and Tree-CNN-20 to 1.8×.
- The initial CIFAR-10 Tree-CNN achieved 98.73% root-node testing accuracy, 86% for the “Animals” branch, 94.43% for “Vehicles,” and 89.10% overall.
Methods
- Tree-CNN is composed of DCNNs in a tree: the root classifies inputs into child nodes, branch nodes perform finer classification, and each leaf node is uniquely associated with one class. Incremental learning samples 10% of new-class training images, computes a softmax likelihood matrix, orders classes, then adds each new class to an existing child, merges child nodes, or adds a new child; affected nodes are retrained with old and new data.
- Experiments used CIFAR-10 (6 initial classes, then 4 new classes) and CIFAR-100 (100 classes, 500 training and 100 testing images per class, randomly divided into 10 groups of 10 and introduced over 9 stages). CIFAR-100 used max depth 2, thresholds 0.1, and maxChildren 5, 10, or 20. Baseline Network B was a VGG-like 11-layer network fine-tuned at five depths (B:I–B:V).
- Metrics were testing accuracy and training effort, defined as the sum over nets of total weights × total training samples. Training used MatConvNet, mini-batch SGD with momentum 0.9, dropout, batch normalization, weight decay 0.001, 300 epochs, and learning rate 0.1 for 200 epochs then reduced by 10 every 50 epochs.
Limitations
- The paper states Tree-CNN continues to grow in size over time, and the implications for memory requirements need investigation; it also notes there is no standardized benchmark protocol for incremental learning, so it used a protocol similar to iCaRL.
- The paper states branch nodes have a higher chance of overfitting as dataset size per node shrinks, so more dropout layers were added; it also notes that if the root node misclassifies an input, the final classification is incorrect regardless of the branch node’s output, and old data must be shown during retraining to avoid catastrophic forgetting.
- Evident from the text: evaluation is limited to CIFAR-10 and CIFAR-100; final CIFAR-100 accuracy for Tree-CNN-5 (61.57%) remains below B:V (63.05%) and below HD-CNN (67.38%) and Hertel et al. (67.68%) as listed in Table 7.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
33 Own interpretation · 22 Own result · 10 Not checked yet · 5 No basis given · 3 Own method
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 size of Network B remains relatively unchanged, as only additional output nodes are added, which translates to a small fraction of new weights in the final layer.”
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.
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, it suffers the worst accuracy degradation over the 9 learning stages (Fig. 6b).”
plainn = 5- 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
“The proposed hierarchical model, when compared against fine-tuning a deep network, achieves significant reduction of training effort, while maintaining competitive accuracy on CIFAR-10 and CIFAR-100.”
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
“At the end of the nine incremental learning stages, certain similar objects grouped together is shown in Fig. 8 for Tree-CNN-10 .”
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.
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 observe that as new classes are added, the Tree-CNNs grow in size by adding more branches (Fig. 5a).”
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.
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 observe that retraining only the FC layers of baseline network ( B:I ) requires the least training effort, however, it gives us the lowest accuracy of 78 : 37% .”
plaincorrelationaln = 5- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The root node achieves a testing accuracy of 98 : 73% , while the branch nodes, 'Animals' and 'Vehicles', achieve 86% and 94 : 43% testing accuracy respectively.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The Tree-CNNs exhibit a lower training effort than 4 fine-tuning test cases, B:II - B:V .”
plaincorrelationaln = 4- 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
“Fig. 6b shows that all the networks suffer from some accuracy degradation with increasing number of classes.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“An interesting thing to note was similar looking classes, that were also semantically similar, were grouped under the same branches.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- It may state two separate claims as one. 23
- Probably not a substantial claim. 12
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
“Tree-CNN-5 almost grows 3 : 4 the size of Network B , while Tree-CNN-10 and Tree-CNN-20 reach 2 : 2 and 1 : 8 the baseline size, respectively.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The root node achieves a testing accuracy of 98 : 73% , while the branch nodes, 'Animals' and 'Vehicles', achieve 86% and 94 : 43% testing accuracy respectively.”
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
“While there were some groups that had object sharing semantic similarity as well, there were odd groups as well, such as Node 13 as shown in Fig. 8.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The three Tree-CNNs perform almost at par with B:IV , and outperform all other variants of network B .”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The Tree-CNN has the second lowest normalized training effort, 40% less than B:V , and 30% less than B:II .”
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
“At the same time, Tree-CNN had comparable accuracy to B:II and B:III , while just being less than the ideal case B:V by a margin of 3 : 76% .”
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.
In the paper
“In recent years Deep Convolutional Neural Networks (DCNNs) have emerged as the leading architecture for large scale image classification [1].”
What the sources say
- Not checked yetcites [1] Aljundi, 2016, Expert gate: Lifelong learning with a network of experts, CoRR
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The initial nodes of the Tree-CNN assign the input into coarse super-classes, and as we approach the leaves of the network, finer classification is done.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“While this example is for a two level Tree-CNN, the algorithm can potentially be extended to deeper Tree-CNN models.”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
None of the sources cited for this claim have been checked yet.
In the paper
“However, in these works, the dataset is fixed from the beginning, and prior knowledge of all the classes and their properties is used to build a hierarchical model.”
What the sources say
- Not checked yetcites [22] Sarwar, S. S., Ankit, A., & Roy, K. (2017). Incremental learning in deep convolutional neural networks using partial network sharing, arXiv preprint arXiv:1712.02719.
- Not checked yetcites [23] Sarwar, 2017, Gabor Filter assisted energy efficient fast learning convolutional neural networks
- Not checked yetcites [24] Shmelkov, 2017, Incremental learning of object detectors without catastrophic forgetting
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The decision on how to grow the tree is semi-supervised: the algorithm itself decides how to grow the tree, given the constraints by the user.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Each leaf node is uniquely associated to a class and no two leaf nodes have the same class.”
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: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds. It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Images belonging to the older dataset are required during retraining, but by localizing the change to a small section of the whole network, our method tries to reduce the training effort and complexity.”
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
It may state two separate claims as one.
In the paper
“It has been observed that initial layers of a CNN learn very generic features [17] that has been exploited for transfer learning [18]' [19].”
What the sources say
- Not checked yetcites [17] Panda, 2017, Semantic driven hierarchical learning for energy-efficient image classification
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“This shows that only retraining the final linear classifier, i.e. the fully connected layer, is not sufficient. We need to train the feature extractors, i.e. convolutional blocks, as well on the new data.”
plaindirectn = 5- 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
“Thus, it can learn with lower training effort than fine-tuning a deep network, while preserving much of the accuracy.”
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.
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
“To compare against the proposed Tree-CNN , we defined a baseline network (Network B ) with a complexity level similar to two stage Tree-CNN .”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Such a method of training and classification can be used to hierarchically classify large datasets.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Once the new classes are allotted locations in the tree, supervised gradient descent based training is performed on the modified/new nodes. This saves us from modifying the whole network, and only affected portions of the network require retraining/fine-tuning.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The final leaf nodes, and the distance between them can also be used as a measure of how similar any two images are.”
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: fails — The paper may give no reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. The paper may give no reason why it holds.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The network organizes the incrementally available data into feature-driven superclasses and improves upon existing hierarchical CNN models by adding the capability of self-growth.”
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
“In 'iCarl' [16], a ResNet-32 is retrained with new data and only 2000 samples of old data (called 'exemplars') at every stage. It is able to recover a good amount performance, compared to 'LwF' but still falls short of state-of-the-art by 18%.”
What the sources say
- Not checked yetcites [16] Panda, 2017, FALCON: Feature driven selective classification for energy-efficient image recognition, IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems source
plaindirectn = 2,000- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The hierarchical node-based learning model of the Tree-CNN attempts to confine the change in the model to a few nodes only. And, in this way, it limits the computation costs of retraining, while using all the previous data.”
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
“Such hierarchical CNN models have been shown to perform at par or even better than standard DCNNs [22].”
What the sources say
- Not checked yetcites [22] Sarwar, S. S., Ankit, A., & Roy, K. (2017). Incremental learning in deep convolutional neural networks using partial network sharing, arXiv preprint arXiv:1712.02719.
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
“Every class is finally associated with a unique leaf node, hence leaf nodes do not require a look up table.”
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.
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
“Training Effort attempts to capture the number of weight updates that happen per training epoch.”
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
“For 'LwF', a ResNet-32 [32] is retrained exclusively on new data at every stage. Hence it suffers the most accuracy degradation.”
What the sources say
- Not checked yetcites [32] Yosinski, 2014, How transferable are features in deep neural networks?
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
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 network grows in a tree-like fashion to accommodate new classes of data, while preserving the ability to distinguish the previously trained classes.”
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In our work, Tree-CNN starts out as a single root node and generates new hierarchies to accommodate the new classes.”
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
“At every incremental learning stage, the root node is trained on all the available data as it needs to learn to classify all the objects into the new branches.”
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
“The Tree-CNN generates hierarchical grouping of initially unrelated classes, thereby generating a label relation graph out of these classes[33].”
What the sources say
- Not checked yetcites [33]
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.
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
“As batch size and number of training epochs is kept the same, the product of the number of weights and the number of training samples used can provide us with the measure of the computation cost of a learning stage.”
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: 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
“Inspired from hierarchical classifiers, our proposed model, Tree-CNN is composed of multiple nodes connected in a tree-like manner. Each node (except leaf nodes) has a DCNN which is trained to classify the input to the node into one of it's children.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
None of the sources cited for this claim have been checked yet.
In the paper
“Common features, that are shared between images, have been used previously to build hierarchical classifiers. These features can be grouped semantically, such as in [20], or be feature-driven, such as 'FALCON' [21].”
What the sources say
- Not checked yetcites [20] Russakovsky, 2015, Imagenet large scale visual recognition challenge, International Journal of Computer Vision source
- Not checked yetcites [21] Rusu, A. A., Rabinowitz, N. C., Desjardins, G., Soyer, H., Kirkpatrick, J., & Kavukcuoglu, K., et al. (2016). Progressive neural networks, arXiv preprint arXiv:1606.04671.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
The paper relies on this without support.
In the real world, data is gathered incrementally over time and all information is not available at once.
“However, in real world, we hardly have all the information at once, and data is, instead, gathered incrementally over time.”
The authors state this as an assumption.
The starting model is already trained to recognize a certain number of objects.
“We start with the assumption that we have a model that is already trained to recognize a certain number of objects.”
The authors state this as an assumption.
The system has access to all the data that has been introduced so far during fine-tuning.
“It is assumed that the system has access to all the data that has been introduced so far.”
The paper relies on this without support.
There is an absence of a standardized benchmark protocol for incremental learning.
“There is an absence of standardized benchmark protocol for incremental learning, which led us to use a benchmarking protocol similar to one used in iCaRL [16].”
The paper relies on this without support.
The dataset available to the user has unique labels assigned to each of its object classes.
“The dataset available to the user will have unique labels assigned to each of it's object classes.”
The paper relies on this without support.
Adding new image classes to a network should be easier than retraining the whole network again with all the classes.
“The motivation of this work stems from the idea that subsequent addition of new image classes to a network should be easier than retraining the whole network again with all the classes.”
The paper relies on this without support.
Catastrophic forgetting is an inherent problem in deep neural networks.
“This is due to the inherent problem of 'catastrophic forgetting' in deep neural networks.”
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?
In the citing paper
“Although this method mitigates catastrophic forgetting to a certain extent, it is intuitively a method that consumes more space and is not efficient in training.”
From An Appraisal of Incremental Learning Methods (2020) · cites this paper as [28]
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
§ 5.2. Sequentially adding new classes (CIFAR-100) · p. 19
“The three Tree-CNNs perform almost at par with B:IV , and outperform all other variants of network B .”
§ Appendix C. Full Simulation Results · p. 27
“| 100 | 31.73 | 48.68 | 54.16 | 60.48 | 63.05 | 61.57 | 60.46 | 59.99 |”
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
Very Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148
ImageNet 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
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
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
Girshick, R. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.169 · Cited by 28,105
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
Learning Multiple Layers of Features from Tiny Images
Krizhevsky, A. (2024) · doi:10.57702/zp44cu3g · Cited by 23,889
How transferable are features in deep neural networks?
Yosinski, J., Clune, J., Bengio, Y., & Lipson, H. (2014) · arXiv · doi:10.48550/arxiv.1411.1792 · Cited by 6,061
Cited by
Sorted by citations
Online continual learning in image classification: An empirical survey
Mai, Z., Li, R., Jeong, J., Quispe, D., Kim, H., & Sanner, S. (2021) · Neurocomputing · doi:10.1016/j.neucom.2021.10.021 · Cited by 468
3D Point Cloud Generative Adversarial Network Based on Tree Structured Graph Convolutions
Shu, D., Park, S., & Kwon, J. (2019) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00396 · Cited by 284
Incremental Learning Techniques for Semantic Segmentation
Michieli, U., & Zanuttigh, P. (2019) · IEEE International Conference on Computer Vision Workshops · doi:10.1109/iccvw.2019.00400 · Cited by 274
“Some methods [16, 27, 28, 36] exploit network architectures which grow during the training process.”
MethodAutomated emotion recognition: Current trends and future perspectives
Maithri, M., Raghavendra, U., Gudigar, A., Samanth, J., Barua, P. D., Murugappan, M., Chakole, Y., & Acharya, U. R. (2022) · Computer Methods and Programs in Biomedicine · doi:10.1016/j.cmpb.2022.106646 · Cited by 172
Crop mapping from image time series: Deep learning with multi-scale label hierarchies
Türkoglu, M. Ö., D’Aronco, S., Perich, G., Liebisch, F., Streit, C., Konrad Schindler, B., & Wegner, J. D. (2021) · Remote Sensing of Environment · doi:10.1016/j.rse.2021.112603 · Cited by 194
Explainable skin lesion diagnosis using taxonomies
Barata, C., Celebi, M. E., & Marques, J. S. (2020) · Pattern Recognition · doi:10.1016/j.patcog.2020.107413 · Cited by 153
“Hierarchical CNNs have been used in a multitude of fields, ranging from coarse-fine image classification [23,24] to image captioning [25,26] .”
MethodIntrusion Detection System Based on Fast Hierarchical Deep Convolutional Neural Network
Mendonça, R. V., Teodoro, A. A. M., Rosa, R. L., Saadi, M., Carrillo, D., Nardelli, P. H. J., & Rodríguez, D. Z. (2021) · IEEE Access · doi:10.1109/access.2021.3074664 · Cited by 143
“Tree-CNN is used in [32].”
BackgroundMethodIncremental Learning in Deep Convolutional Neural Networks Using Partial Network Sharing
Sarwar, S. S., Ankit, A., & Roy, K. (2019) · IEEE Access · doi:10.1109/access.2019.2963056 · Cited by 133
“[27] where the network grows in a tree-like manner to accommodate the new classes.”
BackgroundSelf-citationAn Appraisal of Incremental Learning Methods
Luo, Y., Yin, L., Bai, W., & Mao, K. (2020) · Entropy · doi:10.3390/e22111190 · Cited by 120
“A hierarchical network with a tree structure was designed by Roy et al. [28].”
MethodResultInfluentialAndresini, G., Pendlebury, F., Pierazzi, F., Loglisci, C. A. A. M. D., Appice, A., & Cavallaro, L. (2021) · ACM Workshop on Artificial Intelligence and Security (AISec) · doi:10.1145/3474369.3486864 · Cited by 117
“In the presence of zero-day attacks, the past model may be structurally extended to incorporate new model branches [55].”
BackgroundZhao, D., Cai, W., & Cui, L. (2024) · Advanced Engineering Informatics · doi:10.1016/j.aei.2024.102559 · Cited by 116
Semantics for Robotic Mapping, Perception and Interaction: A Survey
Garg, S., Sünderhauf, N., Dayoub, F., Morrison, D., Cosgun, A., Carneiro, G., Wu, Q., Chin, T.-J., Reid, I. D., Gould, S. J., et al. (2020) · Foundations and Trends in Robotics · doi:10.1561/2300000059 · Cited by 122
Knowledge distillation for incremental learning in semantic segmentation
Michieli, U., & Zanuttigh, P. (2021) · Computer Vision and Image Understanding · doi:10.1016/j.cviu.2021.103167 · Cited by 100
Cui, L., Dong, Z., Xu, H., & Zhao, D. (2023) · Advanced Engineering Informatics · doi:10.1016/j.aei.2023.102322 · Cited by 93
Dong, Z., Jiang, Y., Jiao, W., Zhang, F., Wang, Z., Huang, J., Wang, X., & Zhang, K. (2025) · Advanced Engineering Informatics · doi:10.1016/j.aei.2024.103004 · Cited by 91
Kumar, A., & Patel, V. K. (2023) · Multimedia Tools and Applications · doi:10.1007/s11042-023-14663-z · Cited by 71
“Roy, et al [26] proposed a DCNN-based Hierarchical network structure that adapts that can expand when new data are presented.”
BackgroundMethodContinual Learning: A Review of Techniques, Challenges, and Future Directions
Wickramasinghe, B., Saha, G., & Roy, K. (2023) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2023.3339091 · Cited by 82
“Tree-CNN is a similar tree-like architecture that grows hierarchically to accommodate new classes [101].”
BackgroundSelf-citationEfficient federated learning for fault diagnosis in industrial cloud-edge computing
Wang, Q., Li, Q., Wang, K., Wang, H., & Zeng, P. (2021) · Computing · doi:10.1007/s00607-021-00970-6 · Cited by 66
Early Detection of Network Intrusions Using a GAN-Based One-Class Classifier
Kim, T., & Pak, W. (2022) · IEEE Access · doi:10.1109/access.2022.3221400 · Cited by 72
Incremental multi-view spectral clustering with sparse and connected graph learning
Yin, H., Hu, W., Zhang, Z., Lou, J., & Miao, M. (2021) · Neural Networks · doi:10.1016/j.neunet.2021.08.031 · Cited by 58
AirNet: predictive machine learning model for air quality forecasting using web interface
Rahman, M. M., Nayeem, M., Ahmed, M. S., Tanha, K. A., Sakib, M. S. A., Uddin, K. M. M., & Babu, H. M. H. (2024) · ENVIRONMENTAL SYSTEMS RESEARCH · doi:10.1186/s40068-024-00378-z · Cited by 58
Identifying Reionization Sources from 21cm Maps using Convolutional Neural Networks
Hassan, S., Liu, A., Kohn, S. A., & Plante, P. L. (2018) · Monthly Notices of the Royal Astronomical Society · doi:10.1093/mnras/sty3282 · Cited by 54
Catastrophic Forgetting in Deep Learning: A Comprehensive Taxonomy
Aleixo, E. L., Colonna, J. G., Cristo, M., & Fernandes, E. (2023) · arXiv · doi:10.48550/arxiv.2312.10549 · Cited by 54
Rong, Y., Chen, Q., Fu, Y., Yang, X., Al‐Hallaq, H., Wu, Q., Yuan, L., Xiao, Y., Cai, B., Latifi, K., et al. (2023) · International Journal of Radiation Oncology*Biology*Physics · doi:10.1016/j.ijrobp.2023.10.033 · Cited by 58
Catastrophic Forgetting in Deep Learning: A Comprehensive Taxonomy
Aleixo, E. L., Colonna, J. G., Cristo, M., & Fernandes, E. (2024) · Journal of the Brazilian Computer Society · doi:10.5753/jbcs.2024.3966 · Cited by 54
Kalaiselvan, K., Saravanan, R., Adhavan, B., & Manikandan, G. S. (2023) · Electrical Engineering · doi:10.1007/s00202-023-02095-9 · Cited by 47
“The neural network uses learning and correlation patterns from training input data samples to calculate the usable data [49–52].”
Diversity-driven knowledge distillation for financial trading using Deep Reinforcement Learning
Tsantekidis, A., Passalis, N., & Tefas, A. (2021) · Neural Networks · doi:10.1016/j.neunet.2021.02.026 · Cited by 50
Ding, A., Yi, X., Qin, Y., & Wang, B. (2023) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2023.107382 · Cited by 44
Sirshar, M., Hassan, T., Akram, M. U., & Khan, S. A. (2021) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2021.104435 · Cited by 42
“[65] presented hierarchically-fashioned CNN architecture that is incrementally trained to perform various classification tasks with minimal training efforts.”
Quantum Key Distribution Protocol Selector Based on Machine Learning for Next-Generation Networks
Okey, O. D., Maidin, S. S., Rosa, R. L., Toor, W. T., Carrillo, D., Wuttisittikulkij, L., Saadi, M., & Rodríguez, D. Z. (2022) · Sustainability · doi:10.3390/su142315901 · Cited by 44
Kavitha, N., Soundar, K. R., Karthick, R. V., & Kohila, J. (2024) · Computing · doi:10.1007/s00607-024-01334-6 · Cited by 40
T-DFNN: An Incremental Learning Algorithm for Intrusion Detection Systems
Data, M., & Aritsugi, M. (2021) · IEEE Access · doi:10.1109/access.2021.3127985 · Cited by 40
“Unlike Roy et al.’s approach [25], we did not group similar intrusion classes into one superclass.”
MethodDeep Continual Learning for Emerging Emotion Recognition
Thuseethan, S., Rajasegarar, S., & Yearwood, J. L. (2021) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2021.3116434 · Cited by 41
“The adaptability to work with deep networks and requiring only new task data to train the network are the two notable improvements of LwF. 4) Tree-CNN is proposed in [31], which adjusts the network in a tree-like manner while observing new class data.”
BackgroundMethodWang, X., Gu, H., Wang, T., We, Z., Li, A., & Chu, F. (2021) · Frontiers of Mechanical Engineering · doi:10.1007/s11465-021-0650-6 · Cited by 39
Deep hiearchical multi-label classification applied to chest X-ray abnormality taxonomies
Chen, H., Miao, S., Xu, D., Hager, G. D., & Harrison, A. P. (2020) · Medical Image Analysis · doi:10.1016/j.media.2020.101811 · Cited by 39
Incremental Learning Using a Grow-and-Prune Paradigm With Efficient Neural Networks
Dai, X., Yin, H., & Jha, N. K. (2020) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2020.3037052 · Cited by 36
Semantic clustering based deduction learning for image recognition and classification
Ma, W., Tu, X., Luo, B., & Wang, G. (2021) · Pattern Recognition · doi:10.1016/j.patcog.2021.108440 · Cited by 35
“Tree-CNN [24], proposes training root network by general classes and then learning the fine classes by corresponding growth-network (mainly learned by leaf structure of the network).”
BackgroundInfluentialRamkumar, M., Lakshmi, A., Rajasekaran, M. P., & Manjunathan, A. (2022) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2022.103639 · Cited by 35
Domain-incremental learning for fire detection in space-air-ground integrated observation network
Wang, M., Yu, D., He, W., Yue, P., & Liang, Z. (2023) · International Journal of Applied Earth Observation and Geoinformation · doi:10.1016/j.jag.2023.103279 · Cited by 33
Tamou, A. B., Benzinou, A., & Nasreddine, K. (2022) · Journal of Imaging · doi:10.3390/jimaging8080214 · Cited by 35
“Network Architecture The architecture of our proposed approach is composed of multiple nodes connected in a treelike manner [46] (Figure 5).”
MethodKarthikeyan, R., Balamurugan, V., Cyriac, R., & Sundaravadivazhagan, B. (2022) · Transactions on Emerging Telecommunications Technologies · doi:10.1002/ett.4652 · Cited by 29
“31 • Generally, the tree hierarchical deep convolutional neural network (T-CNN) not reveals any adoption of optimization techniques for computing the optimal parameters for ensuring accurate prediction.”
BackgroundDeep learning‐based automatic classification of three‐level surface information in bridge inspection
Zhang, H., Shen, Z., Lin, Z., Quan, L., & Sun, L. (2023) · Computer-Aided Civil and Infrastructure Engineering · doi:10.1111/mice.13117 · Cited by 30
“Y. Liu et al. (2018) developed visual tree-CNN (VT-CNN) for addressing confusion in categorization, while Roy et al. (2020) proposed Tree-CNN to address catastrophic forgetting.”
MethodResearch on image recognition of intangible cultural heritage based on CNN and wireless network
Liu, E. (2020) · EURASIP Journal on Wireless Communications and Networking · doi:10.1186/s13638-020-01859-2 · Cited by 29
“Local color features are another method of describing color features considering the color space structure of the image [21, 22].”
MethodSelf-Supervised Training Enhances Online Continual LearningPreprint
Gallardo, J., Hayes, T. L., & Kanan, C. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.14010 · Cited by 28
“There are three main approaches to mitigating catastrophic forgetting [53]: 1) increasing model capacity to include new representations that will not interfere with old knowledge [67, 60, 58, 3, 57, 54, 46, 45, 65], 2) regularizing continual parameter updates such that parameters do not deviate too much from their pas…”
BackgroundExtracting and inserting knowledge into stacked denoising auto-encoders
Yu, J., & Liu, G. (2021) · Neural Networks · doi:10.1016/j.neunet.2021.01.010 · Cited by 27
Intra-subject class-incremental deep learning approach for EEG-based imagined speech recognition
García‐Salinas, J. S., Torres-García, A. A., Reyes-Garcia, C. A., & Villaseñor-Pineda, L. (2022) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2022.104433 · Cited by 27
Yan, J., Nuertayi, A., Yan, Y., Liu, S., & Liu, Y. (2023) · Renewable Energy · doi:10.1016/j.renene.2023.118958 · Cited by 27
Li, D., Liu, S., Gao, F., & Sun, X. (2020) · Knowledge-Based Systems · doi:10.1016/j.knosys.2020.106673 · Cited by 26
A Dual-Channel Collaborative Transformer for continual learning
Cai, H., Wang, Y., Luo, Y., & Mao, K. (2025) · Applied Soft Computing · doi:10.1016/j.asoc.2025.112792 · Cited by 26
Modular Neural Networks for Low-Power Image Classification on Embedded Devices
Goel, A., Aghajanzadeh, S., Tung, C., Chen, S.-H., Thiruvathukal, G. K., & Lu, Y.-H. (2020) · ACM Transactions on Design Automation of Electronic Systems · doi:10.1145/3408062 · Cited by 28
“Tree-CNN [66] is an incremental learning technique to train DNNs on previously unseen categories without the requirement of extensive retraining.”
MethodInfluentialChui, K. T., Arya, V., Band, S. S., Alhalabi, M. W., Liu, R. W., & Chi, H. R. (2023) · Journal of Innovation & Knowledge · doi:10.1016/j.jik.2023.100313 · Cited by 26
Digital taxonomist: Identifying plant species in community scientists’ photographs
de Lutio, R., She, Y., D’Aronco, S., Russo, S., Brun, P., Wegner, J. D., & Schindler, K. (2021) · ISPRS Journal of Photogrammetry and Remote Sensing · doi:10.1016/j.isprsjprs.2021.10.002 · Cited by 23
“, 2014), and more recently (Roy et al., 2018) analysed the use of hierarchical labels for visual recognition for the specific case of incremental learning.”
BackgroundChen, Q., Huang, J., Salehi, H. S., Zhu, H., Lian, L., Lai, X., & Wei, K. (2021) · Computer Methods and Programs in Biomedicine · doi:10.1016/j.cmpb.2021.106295 · Cited by 23
Continual Learning Objective for Analyzing Complex Knowledge Representations
Khan, A. M., Hassan, T., Akram, M. U., Alghamdi, N. S., & Werghi, N. (2022) · Sensors · doi:10.3390/s22041667 · Cited by 23
“To overcome such limitations, researchers have proposed different knowledge distillation strategies [3], end-to-end learning [4], learning without forgetting [5], expert gating [6], indefinitely long-term learning (iCaRL) [7], and hierarchical incremental learning [8].”
BackgroundMine-Distill-Prototypes for Complete Few-Shot Class-Incremental Learning in Image Classification
Tai, Y., Tan, Y., Xiong, S., & Tian, J. (2023) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2023.3248601 · Cited by 23
“Roy et al. [26] proposed the adaptive network based on the features hierarchy reorganization.”
BackgroundSGNet: A Super-class Guided Network for Image Classification and Object Detection
Li, K., Wang, N. Y., Yang, Y., & Wang, G. (2021) · Canadian Conference on Computer and Robot Vision · doi:10.1109/crv52889.2021.00025 · Cited by 23
Meenalochini, P., Priya, R. A., Pugalenthi, R., & Jagadeeshwaran, A. (2024) · Journal of Energy Storage · doi:10.1016/j.est.2023.110255 · Cited by 22
Traffic analysis for 5G network slice based on machine learning
Xie, F., Wei, D., & Wang, Z. (2021) · EURASIP Journal on Wireless Communications and Networking · doi:10.1186/s13638-021-01991-7 · Cited by 21
“Traffic mining analysis technology is a technology that captures network traffic and continuously improves the identification algorithm and extracts traffic characteristics according to the changes of network environment [26, 27].”
BackgroundEfficient Computer Vision on Edge Devices with Pipeline-Parallel Hierarchical Neural Networks
Goel, A., Tung, C., Hu, X., Thiruvathukal, G. K., Davis, J. C., & Lu, Y.-H. (2022) · Asia and South Pacific Design Automation Conference (ASP-DAC) · doi:10.1109/asp-dac52403.2022.9712574 · Cited by 21
“We observe that the recent hierarchical DNN architecture [10, 9] is well suited for pipeline parallelism.”
BackgroundLeveraging Class Hierarchies with Metric-Guided Prototype LearningPreprint
Garnot, V. S. F., & Landrieu, L. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2007.03047 · Cited by 29
“Hierarchical Priors: The idea of exploiting the latent taxonomic structure of semantic classes to improve the accuracy of a model has been extensively explored [44], from traditional Bayesian modeling [14, Chapter 5] to adaptive deep learning architectures [2, 38, 41, 50].”
BackgroundKassani, P. H., Lu, F., Guen, Y. L., Belloy, M. E., & He, Z. (2022) · Nature Machine Intelligence · doi:10.1038/s42256-022-00525-0 · Cited by 22
Park, C. H., Kim, H., Lee, J., Ahn, G., Youn, M., & Youn, B. D. (2020) · International Journal of Precision Engineering and Manufacturing-Green Technology · doi:10.1007/s40684-020-00279-3 · Cited by 20
A deep learning method and device for bone marrow imaging cell detection
Liu, J., Yuan, R., Li, Y., Zhou, L., Zhang, Z., Yang, J. J., & Li, X. (2022) · Annals of Translational Medicine · doi:10.21037/atm-22-486 · Cited by 18
IDT: An incremental deep tree framework for biological image classification
Mousser, W., Ouadfel, S., Ahmed, A. T., & Kitouni, I. (2022) · Artificial Intelligence in Medicine · doi:10.1016/j.artmed.2022.102392 · Cited by 19
Review of plant leaf recognition
Wang, Z., Cui, J. H., & Zhu, Y. (2022) · Artificial Intelligence Review · doi:10.1007/s10462-022-10278-2 · Cited by 19
“Roy et al. (2018) proposed an adaptive hierarchical network structure composed of DCNNs, which can grow and learn with the emergence of new data.”
BackgroundTree-structured neural networks: Spatiotemporal dynamics and optimal control
He, J., Xiao, M., Zhao, J., Wang, Z., Yao, Y., & Cao, J. (2023) · Neural Networks · doi:10.1016/j.neunet.2023.04.039 · Cited by 19
Abdullah, A., Fatima, M., Janjua, J. I., & Shabbir, M. (2025) · Systems and Information Engineering Design Symposium · doi:10.1109/sieds65500.2025.11021096 · Cited by 18
“In [21], the author describes a method of using the DCNN-based hierarchical network structure that can expand as new data are introduced.”
Fine-grained recognition: Multi-granularity labels and category similarity matrix
Shu, X., Zhang, L., Wang, Z., Wang, L., & Zhang, Y. (2023) · Knowledge-Based Systems · doi:10.1016/j.knosys.2023.110599 · Cited by 18
New Insights on Relieving Task-Recency Bias for Online Class Incremental Learning
Liang, G., Chen, Z., Chen, Z., Ji, S., & Zhang, Y. (2023) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2023.3325651 · Cited by 20
“They could be parted into dynamic architecture [28], [29], [30] and fixed architecture [14], [31], [32].”
BackgroundSPACE: Structured Compression and Sharing of Representational Space for Continual Learning
Saha, G., Garg, I., Ankit, A., & Roy, K. (2021) · IEEE Access · doi:10.1109/access.2021.3126027 · Cited by 22
“However, in the context of continual learning [4, 5, 6, 7], where humans can continuously adapt and learn new tasks throughout their lifetimes without forgetting the old ones, current deep learning methods struggle to perform well since the data distribution changes over the course of learning.”
BackgroundSelf-citationSpatio-Temporal Event Forecasting Using Incremental Multi-Source Feature Learning
Zhao, L., Gao, Y., Ye, J., Chen, F., Ye, Y., Lu, C., & Ramakrishnan, N. (2021) · ACM Transactions on Knowledge Discovery from Data · doi:10.1145/3464976 · Cited by 18
“For example, Roy et al. [42] develop an adaptive hierarchical network structure composed of deep convolutional neural networks that can grow and learn as new data becomes available.”
Han, Y.-N., & Liu, J. (2022) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2022.104966 · Cited by 17
“[40] propose a hierarchical network based on a tree structure for continual learning.”
BackgroundLee, J.-H., & Sull, S. (2019) · Remote Sensing · doi:10.3390/rs11192276 · Cited by 15
“While the tree structure was also used in option pricing [38], Tree-CNN [39] and the classification and regression tree (CART) [40], our regression tree CNN estimates the GSD by computing the exponent (classification), calculating the mantissa (regression) and then combining them.”
MethodDeep convolutional tree networks
Amory, A. A., Muhammad, G., & Mathkour, H. I. (2019) · Future Generation Computer Systems · doi:10.1016/j.future.2019.06.010 · Cited by 15
A-iLearn: An adaptive incremental learning model for spoof fingerprint detection
Agarwal, S., Rattani, A., & Chowdary, C. R. (2021) · Machine Learning with Applications · doi:10.1016/j.mlwa.2021.100210 · Cited by 15
Image classification using convolutional neural network tree ensembles
Hafiz, A. M., Bhat, R., & Hassaballah, M. (2022) · Multimedia Tools and Applications · doi:10.1007/s11042-022-13604-6 · Cited by 15
“It should also be noted that some works in literature claim to be CNN-Trees [17, 38] but they are mostly complex CNN architectures.”
BackgroundOnline Task-free Continual Learning with Dynamic Sparse Distributed Memory
Pourcel, J., Vu, N., & French, R. M. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19806-9_42 · Cited by 20
“Other methods of this kind dynamically add new layers to the network [51].”
MethodAnytime inference with distilled hierarchical neural ensembles
Ruiz, A., & Verbeek, J. J. (2021) · UPCommons institutional repository (Universitat Politècnica de Catalunya) · doi:10.48550/arxiv.2003.01474 · Cited by 15
“…have been explored for different purposes, such as learning mixture of experts (Liu et al., 2019; Tanno et al., 2019), incremental learning (Roy et al., 2020) or efficient ensemble implementation (Zhang et al., 2018; Lee et al., 2015), our work is the first to leverage this structure for…”
BackgroundA continual learning framework for adaptive defect classification and inspection
Sun, W., Kontar, R. A., Jin, J. (., & Chang, T. (2023) · Journal of Quality Technology · doi:10.1080/00224065.2023.2224974 · Cited by 14
“For example, Xiao et al. (2014) and Roy, Panda, and Roy (2020) propose two hierarchical architectures to allow deep neural networks grow sequentially when new classes of samples arrive.”
BackgroundAutomatic Classification of Equivalent Mutants in Mutation Testing of Android Applications
Kusharki, M. B., Misra, S., Muhammad-Bello, B. L., Salihu, I. A., & Suri, B. (2022) · Symmetry · doi:10.3390/sym14040820 · Cited by 14
“Our research methodology was inspired by the research of [22] but with some modifications to accommodate our contribution in using a standardized dataset as recommended by the work of [10], based on the FAIR principle.”
BackgroundMethodInfluentialMohammed, M., Rahmani, M. K. I., Ahmed, M. E., Irshad, R. R., Yasmin, S., Ahmad, S., Mishra, S., Asopa, P., & Islam, A. (2023) · Journal of Nanoelectronics and Optoelectronics · doi:10.1166/jno.2023.3396 · Cited by 14
A crowdsourcing-based incremental learning framework for automated essays scoring
Bai, H., & Hui, S. C. (2023) · Expert Systems with Applications · doi:10.1016/j.eswa.2023.121755 · Cited by 14
Automatically classifying crime scene images using machine learning methodologies
Abraham, J., Ng, R., Morelato, M., Tahtouh, M., & Roux, C. P. (2021) · Forensic Science International Digital Investigation · doi:10.1016/j.fsidi.2021.301273 · Cited by 14
FoCL: Feature-oriented continual learning for generative models
Lao, Q., Mortazavi, M., Tahaei, M. S., Dutil, F., Fevens, T., & Havaei, M. (2021) · Pattern Recognition · doi:10.1016/j.patcog.2021.108127 · Cited by 14
A consistency-aware deep capsule network for hierarchical multi-label image classification
Noor, K. T., Robles‐Kelly, A., Zhang, L. Y., Bouadjenek, M. R., & Luo, W. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.128376 · Cited by 13
Parallel Multistage Wide Neural Network
Xi, J., Ersoy, O. K., Fang, J., Wu, T., Wei, X., & Zhao, C. (2021) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2021.3120331 · Cited by 13
“Roy et al. [13] proposed a Tree-CNN, which is a hierarchical deep neural network growing in a tree-like manner for lifelong learning.”
BackgroundHierarchical bilinear convolutional neural network for image classification
Zhang, X., Tang, L., Luo, H., Zhong, S., Guan, Z., Chen, L. F., Zhao, C., Peng, J., & Jianping Fan, B. (2021) · IET Computer Vision · doi:10.1049/cvi2.12023 · Cited by 13
“In this subsection, we compare our method with eleven representative image classification methods: VGG16 [2], ResNet ‐ 101 [29], HD ‐ CNN [14], Branch ‐ CNN [8], Coare ‐ to ‐ fine CNN [20], Tree ‐ CNN [30], VI ‐ CNN B [15], B ‐ CNN [7], Highway Net [31], Capsule Net [26] and Autonomous [32].”
MethodHulaMove: Using Commodity IMU for Waist Interaction
Xu, X., Li, J., Yuan, T., He, L., Liu, X., Yan, Y., Wang, Y., Shi, Y., Mankoff, J. C., & Dey, A. K. (2021) · CHI Conference on Human Factors in Computing Systems · doi:10.1145/3411764.3445182 · Cited by 14
“Instead, we used a hierarchical tree-CNN [51] which divided the task into four easier complementary binary or ternary classifcation tasks.”
MethodDetecting Malicious JavaScript Using Structure-Based Analysis of Graph Representation
Rozi, M. F., Ban, T., Ozawa, S., Yamada, A., Takahashi, T., Kim, S., & Inoue, D. (2023) · IEEE Access · doi:10.1109/access.2023.3317266 · Cited by 13
DYSON: Dynamic Feature Space Self-Organization for Online Task-Free Class Incremental Learning
He, Y., Chen, Y., Jin, Y., Dong, S., Wei, X., & Gong, Y. (2024) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52733.2024.02241 · Cited by 12
“On the other hand, some architecture-based methods dynamically add new layers to the network to increase the model’s ability to learn new knowledge, such as [38, 48].”
BackgroundMicro-optics assembly for fast axis collimation by means of convolutional neural network
Khachikyan, A., Pippione, G., Sengünes, M. I., Paoletti, R., & Seyfried, M. (2021) · Optics Express · doi:10.1364/oe.433728 · Cited by 12
Learning With Sharing: An Edge-Optimized Incremental Learning Method for Deep Neural Networks
Hussain, M. A., Huang, S.-A., & Tsai, T.-H. (2023) · IEEE Transactions on Emerging Topics in Computing · doi:10.1109/tetc.2022.3210905 · Cited by 11
“Tree-CNN was proposed by Roy et al. [16] which uses a tree-like architecture to grow the network with the arrival of new classes.”
MethodGranata, V., Fusco, R., Coluccino, S., Russo, C., Grassi, F., Tortora, F., Conforti, R., & Caranci, F. (2024) · La radiologia medica · doi:10.1007/s11547-024-01791-1 · Cited by 11
Enabling Large Dynamic Neural Network Training with Learning-based Memory Management
Ren, J., Xu, D., Yang, S., Zhao, J., Li, Z., Navasca, C., Wang, C., Xu, H., & Li, D. (2024) · IEEE International Symposium on High-Performance Computer Architecture (HPCA) · doi:10.1109/hpca57654.2024.00066 · Cited by 14
Li, H., Deng, H., & Zhang, Y. (2022) · Computational Intelligence and Neuroscience · doi:10.1155/2022/4806763 · Cited by 11
“e hidden layer, includingmany levels, is used for data processing [25].”
MethodRetractedMSDetector: A Static PHP Webshell Detection System Based on Deep-Learning
Cheng, B., Guo, Y., Ren, Y., Yang, G., & Xu, G. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-10363-6_11 · Cited by 11
Grammatically Recognizing Images with Tree Convolution
Wang, G., Wang, G., Wang, K., Liang, X., & Lin, L. (2020) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3394486.3403133 · Cited by 11
“…Existing works suggest that introducing structure information into CNNs is beneficial for many machine learning tasks, such as image classification [38], semantic segmentation [27, 63], object detection [45, 46], language processing [35], knowledge discovery and data mining [23, 72], and general…”
BackgroundAn Automated Mammals Detection Based on SSD-Mobile Net
Alsaadi, E. M. T. A., & Abbadi, N. K. E. (2021) · Journal of Physics Conference Series · doi:10.1088/1742-6596/1879/2/022086 · Cited by 10
“Compared to fine-tuning a deep network, the proposed hierarchical model achieved an important reduction in training effort while retaining competitive accuracy on CIFAR-10, and CIFAR-100 [17].”
MethodPradeep, J., Vijayaragavan, M., Krishnakumar, V., & Kumar, B. S. (2024) · Clean Technologies and Environmental Policy · doi:10.1007/s10098-023-02705-x · Cited by 10
“By then, the AHO-THDCNN method performance is executed in MATLAB platform as well as compared with various existing approaches (Zitouni et al. 2022; Roy et al. 2020).”
Robust Detection of Network Intrusion using Tree-based Convolutional Neural Networks
Mishra, S., Dwivedula, R., Kshirsagar, V., & Hota, C. (2020) · Joint International Conference on Data Science & Management of Data (CODS-COMAD) · doi:10.1145/3430984.3431036 · Cited by 10
“and TreeCNN-reduced are variants of CNN models - while the basic version is modelled based on the classifier node used in [19], TreeCNN-reduced is a modified version incorporating EarlyStopping, ModelCheckpoints and weighted loss functions, to deal with the class imbalance present in network datasets, something not se…”
BackgroundMethodInfluential
Metadata
- Authors
- Roy, D., Panda, P., & Roy, K.
- Affiliations
- Purdue University West Lafayette (US)
- Corresponding author
- Roy, Deboleena
- Venue
- Neural Networks
- Publisher
- Elsevier BV
- Published
- 2020-01
- Volume
- 121
- Pages
- 148-160
- ISSN
- 0893-6080 (ISSN-L)
- OpenAlex
- W2804177786
- Semantic Scholar
- e811c76173d4b0e478d3ab5965358a6377dd48ce
- Cited by
- 257
- References
- 37
- Citations by year
- 2026: 12 · 2025: 29 · 2024: 39 · 2023: 44 · 2022: 41 · 2021: 53 · 2020: 30
- FWCI
- 15.11
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Licence
- Locations
- Neural Networks (acceptedVersion)
- PubMed (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- Semiconductor Research Corporation
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Data Stream Mining Techniques
- Keywords
- incremental learning, catastrophic forgetting, hierarchical convolutional neural network, class-incremental learning, adaptive network structure
- MeSH terms
- Deep Learning, Animals, Humans, Pattern Recognition, Automated, Photic Stimulation, Neural Networks, Computer
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 2026-10-08, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-08 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
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
