Learning a Unified Classifier Incrementally via Rebalancing
Abstract
Conventionally, deep neural networks are trained offline, relying on a large dataset prepared in advance. This paradigm is often challenged in real-world applications, e.g. online services that involve continuous streams of incoming data. Recently, incremental learning receives increasing attention, and is considered as a promising solution to the practical challenges mentioned above. However, it has been observed that incremental learning is subject to a fundamental difficulty -- catastrophic forgetting, namely adapting a model to new data often results in severe performance degradation on previous tasks or classes. Our study reveals that the imbalance between previous and new data is a crucial cause to this problem. In this work, we develop a new framework for incrementally learning a unified classifier, e.g. a classifier that treats both old and new classes uniformly. Specifically, we incorporate three components, cosine normalization, less-forget constraint, and inter-class separation, to mitigate the adverse effects of the imbalance. Experiments show that the proposed method can effectively rebalance the training process, thus obtaining superior performance compared to the existing methods. On CIFAR-100 and ImageNet, our method can reduce the classification errors by more than 6% and 13% respectively, under the incremental setting of 10 phases.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
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?
LUCIR CNN can be deployed with or without a memory for the past.
“This version can be deployed with or without a memory for the past.”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [28]
LUCIR NCM is functional only in the presence of a memory, due to the need for exemplars to compute past class-mean.
“This version is functional in presence of a memory only, due to the need for exemplars to compute past class-mean..”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [28]
Class IL algorithms provide the most interesting results in recent works from literature.
“We focus on class IL algorithms which provide the most interesting results in recent works from literature [7, 8, 14, 23, 24, 28, 56, 75].”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [28]
The iCaRL implementation from [28] has significantly lower performance than the two versions of LUCIR.
“The iCaRL implementation from the same paper [28] has significantly lower performance than the two versions of LUCIR .”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [28]
The other methods with strong performance were ScaIL (GIL = -3.7), BiC (GIL = -4.03) and LUCIR CNN (GIL = -4.13).
“The other methods with strong performance were all proposed recently: ScaIL [8] ( GIL = *3 : 7 ), BiC [75] ( GIL = *4 : 03 ) and LUCIR CNN [28] ( GIL = *4 : 13 ).”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [28]
Only a more sophisticated scheme which combines distillation and an inter-class separation component in LUCIR CNN [28] outperforms FT init, but stays way below FT init siw + mc that does not need distillation.
“Only a more sophisticated scheme which combines distillation and an inter-class separation component in LUCIR CNN [28] outperforms FT init , but stays way below FT init siw + mc that does not need distillation.”
From A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks (2020) · cites this paper as [28]
Cites
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 · see citing works
Overcoming catastrophic forgetting in neural networks
Kirkpatrick, J., Pascanu, R., Rabinowitz, N., Veness, J., Desjardins, G., Rusu, A. A., Milan, K., Quan, J., Ramalho, T., Grabska-Barwinska, A., et al. (2017) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1611835114 · Cited by 7,899 · see citing works
DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition
Donahue, J., Jia, Y., Vinyals, O., Hoffman, J., Zhang, N., Tzeng, E., & Darrell, T. (2013) · arXiv · doi:10.48550/arxiv.1310.1531 · Cited by 4,019 · see citing works
iCaRL: Incremental Classifier and Representation Learning
Rebuffi, S.-A., Kolesnikov, A., Sperl, G., & Lampert, C. H. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.587 · Cited by 3,988 · see citing works
Memory Aware Synapses: Learning What (not) to Forget
Aljundi, R., Babiloni, F., Elhoseiny, M., Rohrbach, M., & Tuytelaars, T. (2018) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-01219-9_9 · Cited by 1,590 · see citing works
Low-Shot Learning with Imprinted Weights
Qi, H., Brown, M., & Lowe, D. G. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00610 · Cited by 539 · see citing works
Encoder Based Lifelong Learning
Rannen, A., Aljundi, R., Blaschko, M. B., & Tuytelaars, T. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.148 · Cited by 324 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
A Comprehensive Survey of Continual Learning: Theory, Method and Application
Wang, L., Zhang, X., Su, H., & Zhu, J. (2024) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2024.3367329 · Cited by 1,113 · see citing works
“stability, emergent strategies include renormalization of the old and new task solutions [166], [263], [384], [442], balanced exploitation of the old and new training samples [10], [167], [174], [452], [469], [513], reserving space for subsequent tasks [9], [199], [518], etc.”
BackgroundMethodInfluentialPrototype Augmentation and Self-Supervision for Incremental Learning
Zhu, F., Zhang, X.-Y., Wang, C., Yin, F., & Liu, C.-L. (2021) · 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr46437.2021.00581 · Cited by 427 · see citing works
Topology-Preserving Class-Incremental Learning
Tao, X., Chang, X., Hong, X., Wei, X., & Gong, Y. (2020) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-58529-7_16 · Cited by 162 · see citing works
An Appraisal of Incremental Learning Methods
Luo, Y., Yin, L., Bai, W., & Mao, K. (2020) · Entropy · doi:10.3390/e22111190 · Cited by 120 · see citing works
Replay in Deep Learning: Current Approaches and Missing Biological Elements
Hayes, T. L., Krishnan, G. P., Bazhenov, M., Siegelmann, H. T., Sejnowski, T. J., & Kanan, C. (2021) · Neural Computation · doi:10.1162/neco_a_01433 · Cited by 106 · see citing works
Federated Learning Empowered Real-Time Medical Data Processing Method for Smart Healthcare
Guo, K., Chen, T., Ren, S., Li, N., Hu, M., & Kang, J. (2024) · IEEE/ACM Transactions on Computational Biology and Bioinformatics · doi:10.1109/tcbb.2022.3185395 · Cited by 63 · see citing works
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
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 · see citing works
Continual Egocentric Activity Recognition With Foreseeable-Generalized Visual–IMU Representations
He, C., Cheng, S., Qiu, Z., Xu, L., Meng, F., Wu, Q., & Li, H. (2024) · IEEE Sensors Journal · doi:10.1109/jsen.2024.3371975 · Cited by 11 · see citing works
Agent-Dice: Disentangling Knowledge Updates via Geometric Consensus for Agent Continual Learning
Wu, Z., Lou, X., Ma, X., Li, Y., Liu, W., Zhang, W., Wang, J., & Zhang, Z. (2026) · arXiv · doi:10.48550/arxiv.2601.03641
“…or feature representations (Kirk-patrick et al., 2017; Zenke et al., 2017), approaches that store and replay a subset of previous data (Rebuffi et al., 2017; Hou et al., 2019), and architecture-based strategies that introduce task-specific modules or models (Schwarz et al., 2018; Yan et al., 2021).”
A Comprehensive Study of Class Incremental Learning Algorithms for Visual Tasks
Belouadah, E., Popescu, A., & Kanellos, I. (2020) · arXiv · doi:10.48550/arxiv.2011.01844 · Cited by 235 · see citing works
“[19] present Learning a Unified Classifier Incrementally via Rebalancing (LUCIR), a method which gains a lot of traction.”
BackgroundMethodInfluential
- Authors
- Hou, S., Pan, X., Loy, C. C., Wang, Z., & Lin, D.
- Affiliations
- University of Science and Technology of China (CN)
- Chinese University of Hong Kong (HK)
- Nanyang Technological University (SG)
- Venue
- 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
- Publisher
- IEEE
- Published
- 2019-06
- Pages
- 831-839
- ISSN
- 1063-6919 (ISSN-L)
- OpenAlex
- W2948734064
- Cited by
- 1,195
- References
- 7
- Citations by year
- 2026: 100 · 2025: 219 · 2024: 274 · 2023: 251 · 2022: 168 · 2021: 128 · 2020: 53 · 2019: 2
- FWCI
- 33.57
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings (publishedVersion)
- 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, class imbalance, inter-class separability, online learning, 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
