Research map / Reinforcement learning / Policy gradient and actor-critic
Sample-Efficient and Safe Deep Reinforcement Learning via Reset Deep Ensemble Agents
Woojun Kim, Yongjae Shin, Jongeui Park, Youngchul Sung
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 |
|---|---|---|---|
| Use RDE instead of DQN | DQN · Atari-100k, IQM, replay ratio 2 | IQM: 1.203 -> 1.377 | Improved |
| Use RDE instead of vanilla reset DQN | DQN · Atari-100k, IQM, replay ratio 4 | IQM: 1.351 -> 1.422 | Improved |
| Use RDE instead of DQN | DQN · MiniGrid FourRooms, replay ratio 1 | test return: 0 -> 0.159 | Improved |
| Use RDE instead of vanilla reset | DQN · MiniGrid GoToDoor-8x8, replay ratio 1 | test return: 0.659 -> 0.944 | Improved |
| Set the DMC reset interval to 4×10^5 rather than use vanilla reset at replay ratio 1 | SAC · DMC, replay ratio 1 | normalized IQM: 1.02 -> 1.20 | Improved |
| Increase ensemble size from N=2 to N=4 (ablation) | RDE+DQN · MiniGrid | performance | Improved |
| Remove the reset mechanism, leaving ensemble learning only (ablation) | RDE+DQN · MiniGrid FourRooms | performance | Hurt |
| Use reset-part (last two layers) instead of reset-all (ablation) | RDE+DQN · MiniGrid FourRoom and SimpleCrossing9 | performance | Hurt |
| Use reset-part (last two layers) instead of reset-all (ablation) | RDE+DQN · MiniGrid easier tasks | performance | Improved |
| Use reset-1 (last layer only) instead of reset-all (ablation) | RDE+SAC · DMC acrobot-swingup, fish-swim, humanoid-run, quadruped-run | performance | Hurt |
| Use reset-1 (last layer only) instead of reset-all (ablation) | RDE+SAC · DMC cheetah-run, finger-turn_hard, hopper-hop, swimmer-swimmer15, walker-run | performance | No clear effect |
| Increase vanilla reset frequency to match RDE's number of reset operations (ablation) | SR · MiniGrid | performance | Hurt |
Closest papers by meaning
- The Courage to Stop: Overcoming Sunk Cost Fallacy in Deep Reinforcement Learning (ICML 2025 · similarity 0.87)
- MAD-TD: Model-Augmented Data stabilizes High Update Ratio RL (ICLR 2025 · similarity 0.86)
- Sample-Efficient Multiagent Reinforcement Learning with Reset Replay (ICML 2024 · similarity 0.86)
- Symmetric Replay Training: Enhancing Sample Efficiency in Deep Reinforcement Learning for Combinatorial Optimization (ICML 2024 · similarity 0.86)
- Novelty-Guided Data Reuse for Efficient and Diversified Multi-Agent Reinforcement Learning (AAAI 2025 · similarity 0.85)
- Intelligent Switching for Reset-Free RL (ICLR 2024 · similarity 0.85)
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.
