Research map / Continual learning / Online and replay-based
Gradient Episodic Memory 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 |
|---|---|---|---|
| Increase episodic memory size from 200 to 5,120 (ablation) | GEM · CIFAR-100 | ACC: 0.487 -> 0.654 | Helped |
| Increase episodic memory size from 200 to 5,120 (ablation) | iCARL · CIFAR-100 | ACC: 0.436 -> 0.508 | Helped |
| Use GEM instead of iCARL at same memory | GEM · CIFAR-100 | ACC: 0.508 -> 0.654 | Helped |
| Increase passes per task from 1 to 5 (ablation) | single · MNIST rotations | ACC: 0.53 -> 0.43 | Hurt |
| Increase passes per task from 1 to 5 (ablation) | multimodal · MNIST rotations | ACC: 0.76 -> 0.59 | Hurt |
| Increase passes per task from 1 to 5 (ablation) | EWC · MNIST rotations | ACC: 0.55 -> 0.61 | Helped |
| Increase passes per task from 1 to 5 (ablation) | GEM · MNIST rotations | ACC: 0.86 -> 0.89 | Helped |
| Use GEM instead of EWC at 5 epochs | GEM · MNIST rotations | BWT: -0.11 -> -0.02 | Helped |
| Use GEM instead of EWC | GEM · MNIST rotations (1 epoch) | ACC: 0.55 -> 0.86 | Helped |
| Compare GEM with iid shuffled single model at 5 epochs | GEM · MNIST rotations | ACC: 0.89 -> 0.89 | No clear effect |
| Use GEM instead of EWC in compute | GEM · MNIST permutations | CPU training time (s): 179 -> 77 | Helped |
| Use GEM instead of EWC in compute | GEM · MNIST rotations | CPU training time (s): 169 -> 135 | Helped |
Closest papers by meaning
- Rainbow Memory: Continual Learning With a Memory of Diverse Samples (CVPR 2021 · similarity 0.89)
- Online Fast Adaptation and Knowledge Accumulation (OSAKA): a New Approach to Continual Learning (NeurIPS 2020 · similarity 0.89)
- PaLoRA: Paced Low-Rank Adaptation for Continual Learning (NeurIPS 2026 · similarity 0.88)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.88)
- Learn more, but bother less: parameter efficient continual learning (NeurIPS 2024 · similarity 0.87)
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 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.