Research map / Continual learning / Class-incremental
Online Prototype Learning for Online Continual Learning
Yujie Wei, Jiaxin Ye, Zhizhong Huang, Junping Zhang, Hongming Shan
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add OnPro (online prototype equilibrium + adaptive prototypical feedback) over the best baseline OCM | online class-incremental learning · CIFAR-10 (M=0.1k) | average accuracy: 47.5 -> 57.8 | Improved |
| add online prototype equilibrium (OPE) to the baseline (ablation) | OnPro · CIFAR-10 (M=0.1k) | average accuracy: 52.0 -> 57.8 | Improved |
| add adaptive prototypical feedback (APF) (ablation) | OnPro · CIFAR-100 (M=0.5k) | average accuracy: 19.3 -> 22.7 | Improved |
| replace random mixup with APF-guided sampling (ablation) | OnPro · CIFAR-10 (M=0.1k) | average accuracy: 53.5 -> 57.8 | Improved |
| include both new-class and seen-class prototype losses in OPE (ablation) | OnPro · CIFAR-10 (M=0.1k) | average accuracy: 54.8 / 55.7 -> 57.8 | Improved |
| let seen-class prototype loss include new classes (ablation) | OnPro · CIFAR-100 (M=0.5k) | average accuracy: 20.8 -> 22.7 | Improved |
| add knowledge distillation to ER | ER · CIFAR-10 (M=0.1k) | average accuracy: 19.4 -> 17.0 | Hurt |
| add projection head to OnPro (ablation) | OnPro · CIFAR-10 (M=0.1k) | average accuracy: 56.1 -> 57.8 | Improved |
| increase APF ratio alpha from 0 to 0.25 (ablation) | OnPro · CIFAR-10 (M=0.2k) | average accuracy: 62.9 -> 65.5 | Improved |
| increase APF ratio alpha beyond 0.1 (ablation) | OnPro · CIFAR-100 (M=0.5k) | average accuracy: 22.7 -> 21.1 | Hurt |
| add rotation augmentation to baselines | ER, ASER, DVC · CIFAR-10 | average accuracy: SCR-Rot 35.8 vs SCR 40.2 at M=0.1k; ER-Rot 30.1 vs ER 19.4 | Mixed |
| OnPro vs baselines on average forgetting on TinyImageNet | OnPro · TinyImageNet | average forgetting: OnPro 21.3 vs iCaRL 9.9 and CoPE 11.9 at M=1k | No clear effect |
Closest papers by meaning
- Non-exemplar Online Class-Incremental Continual Learning via Dual-Prototype Self-Augment and Refinement (AAAI 2024 · similarity 0.91)
- Improving Plasticity in Online Continual Learning via Collaborative Learning (CVPR 2024 · similarity 0.90)
- Enhancing Online Continual Learning with Plug-and-Play State Space Model and Class-Conditional Mixture of Discretization (CVPR 2025 · similarity 0.89)
- A Unified and General Framework for Continual Learning (ICLR 2024 · similarity 0.88)
- Layerwise Proximal Replay: A Proximal Point Method for Online Continual Learning (ICML 2024 · similarity 0.88)
- Forgetting, Ignorance or Myopia: Revisiting Key Challenges in Online Continual Learning (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
