Research map / Reinforcement learning / Offline RL
Uncertainty-Based Offline Reinforcement Learning with Diversified Q-Ensemble
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 clipped Q-learning with an N-network ensemble instead of the standard two critics | SAC · D4RL MuJoCo Gym tasks | normalized average return: 84.5 vs 16.2 average | Helped |
| Add ensemble gradient diversification to clipped ensemble actor-critic | SAC-N · D4RL MuJoCo Gym average | normalized average return: 84.5 -> 85.2 | Helped |
| Add ensemble gradient diversification to clipped ensemble actor-critic | SAC-N · D4RL Adroit pen-human | normalized average return: 9.5±1.1 -> 52.1±8.6 | Helped |
| Increase CQL ensemble size from N=2 to N=5 (ablation) | CQL · halfcheetah-medium | normalized average return: 46.9±0.4 -> 47.1±0.3 | Helped |
| Increase CQL ensemble size from N=2 to N=100 (ablation) | CQL · halfcheetah-medium | normalized average return: 46.9±0.4 -> 43.7±0.2 | Hurt |
| Vary CQL ensemble size from N=2 to N=5, 10, 50, or 100 (ablation) | CQL · hopper-medium | normalized average return: 61.9±6.4; 61.6±6.0; 60.1±4.8; 54.3±2.0; 43.7±0.8 | Hurt |
| Vary CQL ensemble size from N=2 to N=5, 10, 50, or 100 (ablation) | CQL · walker2d-medium | normalized average return: 79.5±3.2; 80.8±4.9; 70.9±0.9; 69.4±0.0; 71.3±3.8 | Mixed |
| Add gradient diversification to CQL with eta=0.5, 1.0, or 5.0 (ablation) | CQL · halfcheetah-medium | normalized average return: 46.9±0.4; 46.5±0.4; 47.2±0.1; 47.4±0.5 | No clear effect |
| Add gradient diversification to CQL with eta=0.5, 1.0, or 5.0 (ablation) | CQL · hopper-medium | normalized average return: 61.9±6.4 -> 65.8±11.2; 69.2±8.8; 60.9±3.2 | Helped |
| Add gradient diversification to CQL with eta=0.5, 1.0, or 5.0 (ablation) | CQL · walker2d-medium | normalized average return: 79.5±3.2 -> 82.2±0.6; 80.5±3.1; 82.1±1.2 | Mixed |
| Replace gradient diversification with direct Q-estimate variance regularization at c=50 (ablation) | SAC-N · walker2d-expert | return: 5236 vs 511 | Hurt |
| Replace gradient diversification with direct Q-estimate variance regularization at c=100 or c=200 (ablation) | SAC-N · walker2d-expert | return: 5236 vs 20; 368 | Hurt |
Closest papers by meaning
- Dynamic Uncertainty Estimation for Offline Reinforcement Learning (AAAI 2025 · similarity 0.92)
- Exclusively Penalized Q-learning for Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.91)
- SUMO: Search-Based Uncertainty Estimation for Model-Based Offline Reinforcement Learning (AAAI 2025 · similarity 0.91)
- Q-Distribution guided Q-learning for offline reinforcement learning: Uncertainty penalized Q-value via consistency model (NeurIPS 2024 · similarity 0.90)
- A Perspective of Q-value Estimation on Offline-to-Online Reinforcement Learning (AAAI 2024 · similarity 0.90)
- Selective Uncertainty Propagation in Offline RL (AAAI 2025 · similarity 0.90)
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.