Research map / Reinforcement learning / Offline RL
An Optimistic Perspective on Offline Reinforcement 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 |
|---|---|---|---|
| Train offline REM with randomized convex combinations instead of QR-DQN | Atari agents trained on the DQN Replay Dataset · Stochastic Atari 2600, 60 games | median normalized score: 118.9% -> 123.8% | Helped |
| Use Adam instead of RMSProp for DQN | Offline DQN · Stochastic Atari 2600, 60 games | median normalized score: 83.4% -> 111.9% | Helped |
| Add naive independent Q-value ensembling | Offline DQN · Stochastic Atari 2600, 60 games | median normalized score: 111.9% -> 111.0% | No clear effect |
| Train each ensemble predictor against the average predicted target Q-values | Ensemble-DQN · Stochastic Atari 2600, 60 games | median normalized score: 111.0% -> 112.1% | No clear effect |
| Reduce the replay dataset to 1% of its tuples (ablation) | Offline REM and QR-DQN · Stochastic Atari 2600, 5 games | normalized score | Hurt |
| Use at least 10% of the replay dataset rather than the full dataset (ablation) | Offline REM and QR-DQN · Stochastic Atari 2600, 5 games | normalized score | No clear effect |
| Use only the first 20 million frames of logged DQN experience (ablation) | Offline REM and QR-DQN · Stochastic Atari 2600, 60 games | normalized score | Helped |
| Use separate Q-networks instead of a multi-head Q-network (ablation) | Offline REM · Stochastic Atari 2600, 6 games | evaluation score | Helped |
Closest papers by meaning
- Towards Robust Offline Reinforcement Learning under Diverse Data Corruption (ICLR 2024 · similarity 0.90)
- A Clean Slate for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- Offline-Boosted Actor-Critic: Adaptively Blending Optimal Historical Behaviors in Deep Off-Policy RL (ICML 2024 · similarity 0.90)
- A Perspective of Q-value Estimation on Offline-to-Online Reinforcement Learning (AAAI 2024 · similarity 0.89)
- Fewer May Be Better: Enhancing Offline Reinforcement Learning with Reduced Dataset (ICLR 2025 · similarity 0.89)
- Beyond OOD State Actions: Supported Cross-Domain Offline Reinforcement Learning (AAAI 2024 · similarity 0.89)
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.