Revisiting Distillation and Incremental Classifier Learning
Open accessThe claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
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?
Distillation induces a bias among past classes, which leads to confusion between their predictions.
“The authors of [29] noted that distillation induces a bias among past classes, which leads to confusion between their predictions. This finding is confirmed by the large number of past-past class confusions e . p; p / associated to LwF in Table 7.”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [29]
Section 6.1 (p. 11) explains that distillation can bias predictions against some older classes, causing poor performance on them. Section 6.4 (p. 13) says the classifier struggled on older classes, but the text does not establish that this bias specifically causes confusion between past classes. (LLM, conf=0.84)
Experimental — accuracy not yet measuredHerding-based exemplar selection was reported to provide only marginal improvement compared to random selection in prior work.
“its usefulness was questioned in [14, 29, 37] where it was reported to provide only marginal improvement compared to random selection.”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [29]
The herding mechanism and the nearest-mean-of-exemplars classification seem to matter less in iCaRL.
“The herding mechanism and the nearest-mean-of-exemplars classification seem to matter less.”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [29]
An iCaRL analysis concludes that its most important components are the fixed-size memory and the distillation loss.
“An iCaRL analysis [29] concludes that its most important components are the fixed-size memory and the distillation loss.”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [29]
Knowledge distillation is commonly used to tackle catastrophic forgetting.
“knowledge distillation is commonly used to tackle catastrophic forgetting [14, 25, 29, 32, 48, 54, 85]”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [29]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
The citation lists haven’t been fetched yet
References 31 · Cited by 53
Research works printed but not linked: 12
- [3]3. Wu, Y., Chen, Y., Wang, L., Ye, Y., Liu, Z., Guo, Y., Zhang, Z., Fu, Y.: Incremental classifier learning with generative adversarial networks. CoRR abs/1802.00853 (2018) · arXiv
- [4]4. Hinton, G., Vinyals, O., Dean, J.: Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531 (2015) · arXiv
- [7]7. French, R.M.: Catastrophic interference in connectionist networks: Can it be predicted, can it be prevented? In: Advances in Neural Information Processing Systems. (1994) 1176-1177
- [9]9. Goodfellow, I.J., Mirza, M., Xiao, D., Courville, A., Bengio, Y.: An empirical investigation of catastrophic forgetting in gradient-based neural networks. arXiv preprint arXiv:1312.6211 (2013) · arXiv
- [11]11. Lampert, C.H., Nickisch, H., Harmeling, S.: Attribute-based classification for zeroshot visual object categorization. IEEE Transactions on Pattern Analysis and Machine Intelligence 36 (3) (2014) 453-465
- [12]12. Li, Z., Hoiem, D.: Learning without forgetting. IEEE Transactions on Pattern Analysis and Machine Intelligence (2017)
- [13]13. Lopez-Paz, D., et al.: Gradient episodic memory for continual learning. In: Advances in Neural Information Processing Systems. (2017) 6470-6479
- [17]17. Zenke, F., Poole, B., Ganguli, S.: Improved multitask learning through synaptic intelligence. arXiv preprint arXiv:1703.04200 (2017) · arXiv
- [19]19. Bucilu, C., Caruana, R., Niculescu-Mizil, A.: Model compression. In: Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, ACM (2006) 535-541
- [20]20. Kemker, R., Kanan, C.: Fearnet: Brain-inspired model for incremental learning. arXiv preprint arXiv:1711.10563 (2017) · arXiv
- [21]21. Rusu, A.A., Rabinowitz, N.C., Desjardins, G., Soyer, H., Kirkpatrick, J., Kavukcuoglu, K., Pascanu, R., Hadsell, R.: Progressive neural networks. arXiv preprint arXiv:1606.04671 (2016) · arXiv
- [22]22. Xiao, T., Zhang, J., Yang, K., Peng, Y., Zhang, Z.: Error-driven incremental learning in deep convolutional neural network for large-scale image classification. In: Proceedings of the 22nd ACM international conference on Multimedia, ACM (2014) 177-186
- Authors
- Javed, K., & Shafait, F.
- Affiliations
- National University of Sciences and Technology (PK)
- Corresponding author
- Javed, Khurram
- Venue
- Lecture Notes in Computer Science
- Publisher
- Springer International Publishing
- Published
- 2019
- Pages
- 3-17
- ISSN
- 0302-9743 (ISSN-L)
- 1611-3349
- OpenAlex
- W2850509156
- Cited by
- 53
- References
- 31
- Citations by year
- 2026: 1 · 2025: 1 · 2024: 3 · 2023: 9 · 2022: 5 · 2021: 11 · 2020: 15 · 2019: 8
- FWCI
- 3.73
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- www.springer.com/tdm (tdm)
- Locations
- Lecture notes in computer science (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- 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 › Machine Learning and Data Classification
- Keywords
- catastrophic forgetting, incremental learning, knowledge distillation, classifier bias, continual learning
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.
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
