Research map / Continual learning / Online and replay-based
CoSCL: Cooperation of Small Continual Learners Is Stronger than a Big One
Liyuan Wang, Xingxing Zhang, Qian Li, Jun Zhu, Yi Zhong
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 CoSCL (5 small learners) | EWC · CIFAR-100-SC (A20) | averaged accuracy: 51.74 -> 60.84 | Improved |
| add CoSCL (5 small learners) | EWC · CIFAR-100-RS (A20) | averaged accuracy: 69.18 -> 76.05 | Improved |
| add CoSCL (5 small learners) | MAS · CIFAR-100-SC (A20) | averaged accuracy: 52.18 -> 60.69 | Improved |
| add CoSCL (5 small learners) | EWC · CUB-200-2011 (A10) | averaged accuracy: 42.29 -> 48.53 | Improved |
| add CoSCL (5 small learners) | EWC · Tiny-ImageNet (A10) | averaged accuracy: 48.38 -> 55.10 | Improved |
| use feature ensemble of small learners instead of single learner (ablation) | EWC · CIFAR-100-SC (A20) | averaged accuracy: 51.74 -> 57.69 | Improved |
| use classifier ensemble instead of single learner | EWC · CIFAR-100-SC (A20) | averaged accuracy: 51.74 -> 43.88 | Hurt |
| add ensemble cooperation loss to feature ensemble (ablation) | Feature Ensemble · CIFAR-100-SC (A20) | averaged accuracy: 57.69 -> 59.49 | Improved |
| add task-adaptive gates to feature ensemble (ablation) | Feature Ensemble · CIFAR-100-SC (A20) | averaged accuracy: 57.69 -> 59.85 | Improved |
| add task-adaptive gates to feature ensemble with EC (ablation) | Feature Ensemble + EC · CIFAR-100-SC (A20) | averaged accuracy: 59.49 -> 60.84 | Improved |
| add ensemble cooperation loss to feature ensemble with gates (ablation) | Feature Ensemble + TG · CIFAR-100-SC (A20) | averaged accuracy: 59.85 -> 60.84 | Improved |
| add CoSCL to unsupervised continual learning | SimSiam · CIFAR-100-RS unsupervised CL | averaged accuracy: 41.38 -> 46.33 | Improved |
Closest papers by meaning
- Efficient Continual Learning with Modular Networks and Task-Driven Priors (ICLR 2021 · similarity 0.90)
- Rethinking the Stability-Plasticity Trade-off in Continual Learning from an Architectural Perspective (ICML 2025 · similarity 0.90)
- Continual Learning with Global Alignment (NeurIPS 2024 · similarity 0.89)
- Improving Plasticity in Online Continual Learning via Collaborative Learning (CVPR 2024 · similarity 0.89)
- STAR: Stability-Inducing Weight Perturbation for Continual Learning (ICLR 2025 · similarity 0.89)
- Optimal Task Order for Continual Learning of Multiple Tasks (ICML 2025 · 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.
