Research map / Continual learning / Online and replay-based
Regularizing Second-Order Influences for Continual Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the second-order influence regularizer to vanilla influence-based selection | Vanilla IF · Split CIFAR-10, CIFAR-100, and miniImageNet; class- and task-incremental settings | ACC: 53.07 vs 47.14; 21.15 vs 17.49; 13.63 vs 12.08 | Helped |
| Use the proposed regularized selection method instead of MetaSP | Continual learning with replay · Split CIFAR-100 task-incremental, m=1000 | ACC: 74.27 vs 76.14 | Hurt |
| Add gradient-matching regularization (ablation) | Vanilla IF · Split CIFAR-10 class-incremental, m=500 | ACC: 47.14 -> 49.86 | Helped |
| Add diversity regularization (ablation) | Vanilla IF · Split CIFAR-10 class-incremental, m=500 | ACC: 47.14 -> 50.59 | Helped |
| Add both gradient-matching and diversity regularizers (ablation) | Vanilla IF · Split CIFAR-10 class-incremental, m=500 | ACC: 47.14 -> 51.61 | Helped |
| Use the paper's Hessian-aware second-order regularizer rather than the tested simpler regularizers (ablation) | Influence-based selection · Split CIFAR-10, m=500, class-incremental | ACC: 51.61 -> 53.07 | Helped |
| Add gradient-matching regularization (ablation) | Vanilla IF · Split CIFAR-10 task-incremental, m=500 | ACC: 91.06 -> 92.30 | Helped |
| Add diversity regularization (ablation) | Vanilla IF · Split CIFAR-10 task-incremental, m=500 | ACC: 91.06 -> 90.69 | Hurt |
| Add both gradient-matching and diversity regularizers (ablation) | Vanilla IF · Split CIFAR-10 task-incremental, m=500 | ACC: 91.06 -> 91.93 | Helped |
| Use the paper's Hessian-aware second-order regularizer rather than the tested simpler regularizers (ablation) | Influence-based selection · Split CIFAR-10, m=500, task-incremental | ACC: 92.30 -> 92.53 | Helped |
Closest papers by meaning
- Predicting the Susceptibility of Examples to Catastrophic Forgetting (ICML 2025 · similarity 0.90)
- A Unified and General Framework for Continual Learning (ICLR 2024 · similarity 0.88)
- Summarizing Stream Data for Memory-Constrained Online Continual Learning (AAAI 2024 · similarity 0.87)
- Layerwise Proximal Replay: A Proximal Point Method for Online Continual Learning (ICML 2024 · similarity 0.87)
- The Inter-batch Diversity of Samples in Experience Replay for Continual Learning (AAAI 2024 · similarity 0.87)
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.87)
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.