Research map / Continual learning / Online and replay-based
Elastic Weight Consolidation Done Right for Continual Learning
Xuan Liu, Xiaobin Chang
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 Logits Reversal when computing the Fisher importance (EWC to EWC-DR) | EWC · CIFAR-100 EFCIL big-start T=5 | A_last: 14.61 -> 50.23 | Improved |
| add Logits Reversal when computing the Fisher importance | EWC · ImageNet-Subset EFCIL big-start T=5 | A_last: 11.44 -> 66.18 | Improved |
| add Logits Reversal when computing the Fisher importance | EWC · Tiny-ImageNet EFCIL big-start T=5 | A_last: 9.74 -> 38.24 | Improved |
| add Logits Reversal when computing the Fisher importance | EWC · CIFAR-100 EFCIL equally split T=20 | A_last: 5.48 -> 18.00 | Improved |
| add Logits Reversal when computing the Fisher importance | EWC · MCIT NLVR2 after final task | forgetting transfer (%): 90.66 -> 27.48 | Improved |
| add Logits Reversal when computing the Fisher importance | EWC · MCIT after final task VCR | incremental accuracy A_t (%): 42.99 -> 52.59 | Improved |
| add Logits Reversal when computing the Fisher importance | EWC · MCIT new-task accuracy (plasticity) | diagonal accuracy: VCR 59.30 -> 59.38 | No clear effect |
| accumulate importance online with decay (online EWC) instead of vanilla EWC | EWC · CIFAR-100 EFCIL big-start T=5 | A_last: 14.61 -> 29.70 | Improved |
| use online EWC instead of vanilla EWC in equally-split setting | EWC · ImageNet-Subset EFCIL equally split T=10 | A_last: 10.60 -> 12.78 | Mixed |
| use Synaptic Intelligence instead of vanilla EWC | EWC · ImageNet-Subset EFCIL big-start T=5 | A_last: 11.44 -> 9.19 | Hurt |
| use MAS importance instead of vanilla EWC | EWC · CIFAR-100 EFCIL big-start T=5 | A_last: 14.61 -> 35.37 | Improved |
Closest papers by meaning
- Revisiting Weight Regularization for Low-Rank Continual Learning (ICLR 2026 · similarity 0.93)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.87)
- PaLoRA: Paced Low-Rank Adaptation for Continual Learning (NeurIPS 2026 · similarity 0.87)
- STAR: Stability-Inducing Weight Perturbation for Continual Learning (ICLR 2025 · similarity 0.87)
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.87)
- Large Continual Instruction Assistant (ICML 2025 · 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.
