Research map / Continual learning / Online and replay-based
Does Continual Learning Meet Compositionality? New Benchmarks and An Evaluation Framework
Weiduo Liao, Ying Wei, Mingchen Jiang, Qingfu Zhang, Hisao Ishibuchi
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 |
|---|---|---|---|
| use a modularity-based method (RPSnet) instead of Finetune in class-IL | continual learner, ResNet-18 · COBJ | Acon: 6.37 -> 21.93 | Improved |
| use RPSnet (modularity-based) instead of ER for compositional generalization | continual learner, ResNet-18 · CGQA | Hn: 68.46 -> 59.94 | Hurt |
| use RPSnet instead of ER for compositional generalization | continual learner, ResNet-18 · COBJ | Hn: 45.86 -> 70.47 | Improved |
| use MNTDP* (frozen modules) instead of ER* in task-IL | continual learner, ResNet-18 · CGQA | Hn: 70.19 -> 47.39 | Hurt |
| use LwF instead of Finetune in class-IL | continual learner, ResNet-18 · CGQA | Hn: 63.75 -> 71.03 | Improved |
| use joint multi-task training instead of sequential continual training | ResNet-18 · CGQA | Hn: 71.03 -> 80.35 | Improved |
| use single-head (class-IL) instead of multi-head (task-IL) classifier | Finetune · CGQA | Hn: 69.66 -> 63.75 | Hurt |
| use single-head (class-IL) instead of multi-head (task-IL) classifier | LwF · CGQA | Hn: 69.40 -> 71.03 | No clear effect |
| add experience replay (ER) to Finetune | continual learner, ResNet-18 · CGQA | Acon: 8.38 -> 19.78 | Improved |
| use regularization (GEM, EWC) instead of replay | continual learner, ResNet-18 · CGQA | Hn: GEM 67.33, EWC 66.95 vs Finetune 63.75 | No clear effect |
| test on novel combinations of old concepts vs fresh concepts | continual learners · CGQA | S(sys) relative drop | Hurt |
| increase training samples per class in continual tasks | continual learners · CGQA | Asys and Anon | Improved |
Closest papers by meaning
- Unlocking Compositional Generalization in Continual Few-Shot Learning (NeurIPS 2026 · similarity 0.89)
- Separating the 'what' and 'how' of compositional computation to enable reuse and continual learning (NeurIPS 2025 · similarity 0.89)
- Not Just Object, But State: Compositional Incremental Learning without Forgetting (NeurIPS 2024 · similarity 0.88)
- Learning to Compose: Improving Object Centric Learning by Injecting Compositionality (ICLR 2024 · similarity 0.87)
- Does Data Scaling Lead to Visual Compositional Generalization? (ICML 2025 · similarity 0.87)
- A Probabilistic Framework for Modular Continual Learning (ICLR 2024 · similarity 0.86)
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.
