Research map / Continual learning / Online and replay-based
Experience Replay 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 |
|---|---|---|---|
| Use CLEAR with replay and behavioral cloning instead of standard sequential training | DMLab cyclic task sequence · DMLab | cumulative performance | Helped |
| Use CLEAR instead of Progress & Compress | Atari sequential task sequence · Atari | cumulative performance | Helped |
| Use CLEAR instead of Elastic Weight Consolidation | Atari sequential task sequence · Atari | cumulative performance | Helped |
| Add behavioral cloning losses to replay-based learning (ablation) | CLEAR · DMLab cyclic task sequence | catastrophic forgetting | Helped |
| Use off-policy replay in addition to new on-policy experience (ablation) | CLEAR · DMLab cyclic task sequence | catastrophic forgetting | Helped |
| Use 100% replay instead of a mixture of new and replay experience (ablation) | CLEAR · DMLab cyclic task sequence | overall performance | Hurt |
| Use a 50-50 new-replay mixture instead of 100% replay (ablation) | CLEAR · DMLab cyclic task sequence | overall performance | Helped |
| Use 100% replay instead of a new-replay mixture (ablation) | CLEAR · natlab_varying_map_randomized probe task introduced at different sequence positions | probe-task performance | Hurt |
| Reduce replay-buffer capacity to 5 million frames (ablation) | CLEAR · DMLab sequence with 900 million environmental frames | robustness to catastrophic forgetting | Hurt |
| Use small replay buffers instead of a large buffer (ablation) | CLEAR · DMLab sequence with 900 million environmental frames | performance | No clear effect |
| Train sequentially without replay rather than with replay (ablation) | Standard on-policy baseline · DMLab cyclic task sequence | catastrophic forgetting | Hurt |
| Train with replay and behavioral cloning rather than standard sequential training | CLEAR · DMLab probe task introduced at different sequence positions | probe-task performance | No clear effect |
Closest papers by meaning
- Predicting the Susceptibility of Examples to Catastrophic Forgetting (ICML 2025 · similarity 0.89)
- A Forget-and-Grow Strategy for Deep Reinforcement Learning Scaling in Continuous Control (ICML 2025 · similarity 0.89)
- A Unified and General Framework for Continual Learning (ICLR 2024 · similarity 0.89)
- Overcoming Catastrophic Forgetting in Visual Continual Learning with Reinforcement Fine-Tuning (NeurIPS 2026 · similarity 0.89)
- Mitigating Catastrophic Forgetting in Online Continual Learning by Modeling Previous Task Interrelations via Pareto Optimization (ICML 2024 · similarity 0.89)
- Pareto Continual Learning: Preference-Conditioned Learning and Adaption for Dynamic Stability-Plasticity Trade-off (AAAI 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.