Research map / Reinforcement learning / Policy gradient and actor-critic
Remember and Forget for Experience Replay
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 ReF-ER instead of conventional experience replay | DDPG · MuJoCo tasks | mean cumulative reward | Helped |
| Use ReF-ER instead of conventional experience replay | NAF · MuJoCo tasks | mean cumulative reward | Helped |
| Use ReF-ER instead of conventional experience replay | V-RACER · MuJoCo tasks | mean cumulative reward | Helped |
| Use ReF-ER instead of conventional experience replay | DDPG · Partially observable flow control | mean cumulative reward | Helped |
| Use ReF-ER instead of conventional experience replay | V-RACER · Partially observable flow control | mean cumulative reward | Helped |
| Use rank-based prioritized experience replay instead of conventional replay | DDPG · MuJoCo tasks | mean cumulative reward | Hurt |
| Use rank-based prioritized experience replay instead of conventional replay | NAF · MuJoCo tasks | mean cumulative reward | Hurt |
| Add Rule 1, which zeroes gradients from far-policy samples (ablation) | ReF-ER · DDPG, MuJoCo tasks | mean cumulative reward | Hurt |
| Add Rule 1, which zeroes gradients from far-policy samples, to Rule 2 (ablation) | ReF-ER · DDPG, MuJoCo tasks | mean cumulative reward | Helped |
| Add Rule 1, which zeroes gradients from far-policy samples, to Rule 2 (ablation) | ReF-ER · NAF, MuJoCo tasks | mean cumulative reward | Helped |
| Use Rule 2 alone, penalizing policy divergence from replayed behaviors (ablation) | ReF-ER · V-RACER, MuJoCo tasks | mean cumulative reward | Mixed |
| Use Rule 1 alone, zeroing gradients from far-policy samples (ablation) | ReF-ER · V-RACER, MuJoCo tasks | mean cumulative reward | Hurt |
Closest papers by meaning
- Revisiting Fundamentals of Experience Replay (ICML 2020 · similarity 0.93)
- A Forget-and-Grow Strategy for Deep Reinforcement Learning Scaling in Continuous Control (ICML 2025 · similarity 0.89)
- The Courage to Stop: Overcoming Sunk Cost Fallacy in Deep Reinforcement Learning (ICML 2025 · similarity 0.88)
- When Does Non-Uniform Replay Matter in Reinforcement Learning? (NeurIPS 2026 · similarity 0.88)
- Turning Sand to Gold: Recycling Data to Bridge On-Policy and Off-Policy Learning via Causal Bound (NeurIPS 2025 · similarity 0.88)
- Offline-Boosted Actor-Critic: Adaptively Blending Optimal Historical Behaviors in Deep Off-Policy RL (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.