Abstract
Few-shot class-incremental learning (FSCIL) aims to continually learn new semantics given a few training samples of new classes. As training examples are too few to construct good representation upon, FSCIL is required to generalize learned semantics from old to new classes, as well as reduce the representation aliasing between them (old classes ‘forgetting’). This motivates us to develop overcomplete-to-sparse representation learning (O2SRL). It solves the ‘new class generalization’ and ‘old class forgetting’ problems systematically by regularizing both feature completeness and sparsity. Specifically, O2SRL consists of a spatial excitation module (SEM) and a channel purification module (CPM). SEM drives the model to learn overcomplete and generic features, which not only represent all classes well but also benefit generalization to new classes. CPM regularizes the sparsity and uniqueness of features, reducing semantic aliasing between classes and alleviating the forgetting of old classes. These two modules facilitate each other to configure unique and robust representation for both old and new classes. Experiments show that O2SRL improves the state-of-the-art of FSCIL by significant margins on public datasets including CUB200, CIFAR100, and mini-ImageNet. O2SRL’s effectiveness is also validated under the general few-shot learning setting.
Similar content being viewed by others
Data availability
CUB200 [52]: https://www.vision.caltech.edu/datasets/cub_200_2011/ mini-ImageNet [53]: https://www.kaggle.com/datasets/arjunashok33/miniimagenet CIFAR100: https://www.cs.toronto.edu/~kriz/cifar.html
References
Reboud, A., Harrando, I., Lisena, P., Troncy, R.: Stories of love and violence: zero-shot interesting events’ classification for unsupervised tv series summarization. Multimed. Syst. (2023). https://doi.org/10.1007/S00530-022-01040-3
Greenwald, A.G.: Cognitive learning, cognitive response to persuasion, and attitude change. Psychol. Found. Attitud. (1968)
Zhang, C., Song, N., Lin, G., Zheng, Y., Pan, P., Xu, Y.: Few-shot incremental learning with continually evolved classifiers. In: IEEE CVPR (2021). https://doi.org/10.1109/CVPR46437.2021.01227
Zhou, D.-W., Wang, F.-Y., Ye, H.-J., Ma, L., Pu, S., Zhan, D.-C.: Forward compatible few-shot class-incremental learning. In: IEEE CVPR, pp. 9046–9056 (2022). https://doi.org/10.1109/CVPR52688.2022.00884
Liu, H., Gu, L., Chi, Z., Wang, Y., Yu, Y., Chen, J., Tang, J.: Few-shot class-incremental learning viaentropy-regularized data-free replay. Springer, Cham (2022). https://doi.org/10.1007/978-3-031-20053-3_9
Chatterjee, R., Chatterjee, A., Islam, S.H., Khan, M.K.: An object detection-based few-shot learning approach for multimedia quality assessment. Multimedia Systems, 1–14 (2022). https://doi.org/10.1007/S00530-021-00881-8
Li, Z., He, J., Ni, T., Huo, J.: Numerical computation based few-shot learning for intelligent sea surface temperature prediction. Multimed. Syst. 29(5), 3001–3013 (2022). https://doi.org/10.1007/S00530-022-00941-7
Vinyals, O., Blundell, C., Lillicrap, T., Kavukcuoglu, K., Wierstra, D.: Matching networks for one shot learning. In: NeurIPS, pp. 3630–3638 (2016). https://proceedings.neurips.cc/paper/2016/hash/90e1357833654983612fb05e3ec9148c-Abstract.html
Snell, J., Swersky, K., Zemel, R.S.: Prototypical networks for few-shot learning. In: NeurIPS, pp. 4077–4087 (2017)
Sung, F., Yang, Y., Zhang, L., Xiang, T., Torr, P.H.S., Hospedales, T.M.: Learning to compare: Relation network for few-shot learning. In: IEEE CVPR, pp. 1199–1208 (2018). https://doi.org/10.1109/CVPR.2018.00131
Zhang, C., Cai, Y., Lin, G., Shen, C.: Deepemd: Few-shot image classification with differentiable earth mover’s distance and structured classifiers. In: IEEE CVPR, pp. 12200–12210 (2020). https://doi.org/10.1109/CVPR42600.2020.01222
Liu, B., Jiao, J., Ye, Q.: Harmonic feature activation for few-shot semantic segmentation. IEEE Trans. Image Process. 30, 3142–3153 (2021). https://doi.org/10.1109/TIP.2021.3058512
Yang, B., Wan, F., Liu, C., Li, B., Ji, X., Ye, Q.: Part-based semantic transform for few-shot semantic segmentation. IEEE Trans. Neural Netw. Learn. Syst. (2021). https://doi.org/10.1109/TNNLS.2021.3084252
Zhang, C., Cai, Y., Lin, G., Shen, C.: Deepemd: differentiable earth mover’s distance for few-shot learning. IEEE Transactions on Pattern Analysis and Machine Intelligence (2022)
Finn, C., Abbeel, P., Levine, S.: Model-agnostic meta-learning for fast adaptation of deep networks. In: ICML, pp. 1126–1135 (2017). http://proceedings.mlr.press/v70/finn17a.html
Elsken, T., Staffler, B., Metzen, J.H., Hutter, F.: Meta-learning of neural architectures for few-shot learning. In: IEEE CVPR, pp. 12362–12372 (2020). https://doi.org/10.1109/CVPR42600.2020.01238
Sun, Q., Liu, Y., Chua, T., Schiele, B.: Meta-transfer learning for few-shot learning. In: IEEE CVPR, pp. 403–412 (2019). https://doi.org/10.1109/CVPR.2019.00049
Li, B., Yang, B., Liu, C., Liu, F., Ji, R., Ye, Q.: Beyond max-margin: class margin equilibrium for few-shot object detection. In: IEEE CVPR, pp. 7363–7372 (2021). https://doi.org/10.1109/CVPR46437.2021.00728
Zhang, H., Zhang, J., Koniusz, P.: Few-shot learning via saliency-guided hallucination of samples. In: IEEE ICCV, pp. 2770–2779 (2019). https://doi.org/10.1109/CVPR.2019.00288
Li, K., Zhang, Y., Li, K., Fu, Y.: Adversarial feature hallucination networks for few-shot learning. In: IEEE ICCV, pp. 13467–13476 (2020). https://doi.org/10.1109/CVPR42600.2020.01348
Kim, J., Kim, H., Kim, G.: Model-agnostic boundary-adversarial sampling for test-time generalization in few-shot learning. In: ECCV, pp. 599–617 (2020). https://doi.org/10.1007/978-3-030-58452-8_35
Masana, M., Liu, X., Twardowski, B., Menta, M., Bagdanov, A.D., Weijer, J.: Class-incremental learning: survey and performance evaluation. CoRR (2020). arxiv:2010.15277
Rebuffi, S., Kolesnikov, A., Sperl, G., Lampert, C.H.: icarl: Incremental classifier and representation learning. In: IEEE CVPR, pp. 5533–5542 (2017). https://doi.org/10.1109/CVPR.2017.587
Chaudhry, A., Dokania, P.K., Ajanthan, T., Torr, P.H.S.: Riemannian walk for incremental learning: Understanding forgetting and intransigence. In: ECCV, pp. 556–572 (2018). https://doi.org/10.1007/978-3-030-01252-6_33
Wu, Y., Chen, Y., Wang, L., Ye, Y., Liu, Z., Guo, Y., Fu, Y.: Large scale incremental learning. In: IEEE CVPR, pp. 374–382 (2019). https://doi.org/10.1109/CVPR.2019.00046
Shin, H., Lee, J.K., Kim, J., Kim, J.: Continual learning with deep generative replay. In: NeurIPS, pp. 2990–2999 (2017). https://proceedings.neurips.cc/paper/2017/hash/0efbe98067c6c73dba1250d2beaa81f9-Abstract.html
Xiang, Y., Fu, Y., Ji, P., Huang, H.: Incremental learning using conditional adversarial networks. In: IEEE ICCV, pp. 6618–6627 (2019). https://doi.org/10.1109/ICCV.2019.00672
Kim, C.D., Jeong, J., Moon, S., Kim, G.: Continual learning on noisy data streams via self-purified replay. In: IEEE ICCV, pp. 537–547 (2021). https://doi.org/10.1109/ICCV48922.2021.00058
Smith, J., Hsu, Y.-C., Balloch, J., Shen, Y., Jin, H., Kira, Z.: Always be dreaming: a new approach for data-free class-incremental learning. In: IEEE ICCV, pp. 9374–9384 (2021). https://doi.org/10.1109/ICCV48922.2021.00924
Li, Z., Hoiem, D.: Learning without forgetting. IEEE Trans. Pattern Anal. Mach. Intell. 40(12), 2935–2947 (2018). https://doi.org/10.1109/TPAMI.2017.2773081
Dhar, P., Singh, R.V., Peng, K., Wu, Z., Chellappa, R.: Learning without memorizing. In: IEEE CVPR, pp. 5138–5146 (2019). https://doi.org/10.1109/CVPR.2019.00528
Zenke, F., Poole, B., Ganguli, S.: Continual learning through synaptic intelligence. In: ICML, pp. 3987–3995 (2017). http://proceedings.mlr.press/v70/zenke17a.html
Selvaraju, R.R., Cogswell, M., Das, A., Vedantam, R., Parikh, D., Batra, D.: Grad-cam: visual explanations from deep networks via gradient-based localization. In: IEEE ICCV, pp. 618–626 (2017). https://doi.org/10.1109/ICCV.2017.74
Hou, S., Pan, X., Loy, C.C., Wang, Z., Lin, D.: Learning a unified classifier incrementally via rebalancing. In: IEEE CVPR, pp. 831–839 (2019). https://doi.org/10.1109/CVPR.2019.00092
Hu, X., Tang, K., Miao, C., Hua, X.-S., Zhang, H.: Distilling causal effect of data in class-incremental learning. In: IEEE CVPR, pp. 3957–3966 (2021). https://doi.org/10.1109/CVPR46437.2021.00395
Cha, H., Lee, J., Shin, J.: Co2l: contrastive continual learning. In: IEEE ICCV, pp. 9516–9525 (2021). https://doi.org/10.1109/ICCV48922.2021.00938
Mallya, A., Lazebnik, S.: Packnet: adding multiple tasks to a single network by iterative pruning. In: IEEE CVPR, pp. 7765–7773 (2018). https://doi.org/10.1109/CVPR.2018.00810
Serrà, J., Suris, D., Miron, M., Karatzoglou, A.: Overcoming catastrophic forgetting with hard attention to the task. In: ICML. In: Proceedings of Machine Learning Research, pp. 4555–4564 (2018). http://proceedings.mlr.press/v80/serra18a.html
Yoon, J., Yang, E., Lee, J., Hwang, S.J.: Lifelong learning with dynamically expandable networks. In: ICLR (2018). https://openreview.net/forum?id=Sk7KsfW0-
Mallya, A., Davis, D., Lazebnik, S.: Piggyback: adapting a single network to multiple tasks by learning to mask weights. In: ECCV, pp. 72–88 (2018). https://doi.org/10.1007/978-3-030-01225-0_5
Liu, Y., Su, Y., Liu, A., Schiele, B., Sun, Q.: Mnemonics training: Multi-class incremental learning without forgetting. In: IEEE CVPR, pp. 12242–12251 (2020). https://doi.org/10.1109/CVPR42600.2020.01226
Belouadah, E., Popescu, A.: IL2M: class incremental learning with dual memory. In: IEEE ICCV, pp. 583–592 (2019). https://doi.org/10.1109/ICCV.2019.00067
Shmelkov, K., Schmid, C., Alahari, K.: Incremental learning of object detectors without catastrophic forgetting. In: IEEE ICCV, pp. 3420–3429 (2017). https://doi.org/10.1109/ICCV.2017.368
Riemer, M., Cases, I., Ajemian, R., Liu, M., Rish, I., Tu, Y., Tesauro, G.: Learning to learn without forgetting by maximizing transfer and minimizing interference. In: ICLR (2019). https://openreview.net/forum?id=B1gTShAct7
Tian, S., Li, L., Li, W., Ran, H., Ning, X., Tiwari, P.: A survey on few-shot class-incremental learning (2023). https://doi.org/10.48550/arXiv.2304.08130
Kim, D.-Y., Han, D.-J., Seo, J., Moon, J.: Warping the space: Weight space rotation for class-incremental few-shot learning. In: International Conference on Learning Representations (2023). https://api.semanticscholar.org/CorpusID:259298246
Yang, Y., Yuan, H., Li, X., Lin, Z., Torr, P.H.S., Tao, D.: Neural collapse inspired feature-classifier alignment for few-shot class incremental learning (2023). https://api.semanticscholar.org/CorpusID:256615229. ArXiv:2302.03004
Tao, X., Hong, X., Chang, X., Dong, S., Wei, X., Gong, Y.: Few-shot class-incremental learning. In: IEEE CVPR, pp. 12180–12189 (2020)
Zhu, K., Cao, Y., Zhai, W., Cheng, J., Zha, Z.-J.: Self-promoted prototype refinement for few-shot class-incremental learning. In: IEEE CVPR, pp. 6801–6810 (2021). https://doi.org/10.1109/CVPR46437.2021.00673
Papyan, V., Romano, Y., Elad, M.: Convolutional neural networks analyzed via convolutional sparse coding. J. Mach. Learn. Res. 2887–2938 (2017). http://jmlr.org/papers/v18/16-505.html
Murdock, C., Lucey, S.: Reframing neural networks: deep structure in overcomplete representations. CoRR (2021). arxiv:2103.05804
Wah, C., Branson, S., Welinder, P., Perona, P., Belongie, S. Technical Report CNS-TR-2011-001, California Institute of Technology (2011)
Ravi, S., Larochelle, H.: Optimization as a model for few-shot learning. In: International Conference on Learning Representations (2016). https://api.semanticscholar.org/CorpusID:67413369
Chi, Z., Gu, L., Liu, H., Wang, Y., Yu, Y., Tang, J.: Metafscil: a meta-learning approach for few-shot class incremental learning. In: IEEE CVPR, pp. 14166–14175 (2022). https://doi.org/10.1109/CVPR52688.2022.01377
Zhuang, H., Weng, Z., He, R., Lin, Z., Zeng, Z.: Gkeal: Gaussian kernel embedded analytic learning for few-shot class incremental task. In: IEEE CVPR (2023). https://doi.org/10.1109/CVPR52729.2023.00748
Afrasiyabi, A., Larochelle, H., Lalonde, J.-F., Gagné, C.: Matching feature sets for few-shot image classification. In: IEEE CVPR, pp. 9014–9024 (2022). https://doi.org/10.1109/CVPR52688.2022.00881
Funding
This work was supported by National Natural Science Foundation of China (NSFC) under Grants 62225208 and 62171431.
Author information
Authors and Affiliations
Contributions
Fu Mengying proposed the idea and was responsible for the validation of the main experiment. Liu Binghao wrote the main manuscript text and Ma Tianren made relevant pictures and tables. Ye Qixiang has revised the article and gave guidance. All authors reviewed the manuscript.
Corresponding author
Ethics declarations
Conflict of interest
The authors have no Conflict of interest to declare that are relevant to the content of this article.
Ethics approval and consent to participate
Not applicable.
Additional information
Communicated by J. Gao.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Mengying, F., Binghao, L., Tianren, M. et al. Overcomplete-to-sparse representation learning for few-shot class-incremental learning. Multimedia Systems 30, 102 (2024). https://doi.org/10.1007/s00530-024-01294-z
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00530-024-01294-z