Research map / Reinforcement learning / Visual and representation RL
Mastering Atari Games with Limited Data
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 |
|---|---|---|---|
| add self-supervised temporal consistency loss (ablation) | EfficientZero · Atari 100k (26 games) | normed mean: 0.881 -> 1.943 | Helped |
| add self-supervised temporal consistency loss (ablation) | EfficientZero · Atari 100k (26 games) | normed median: 0.340 -> 1.090 | Helped |
| add end-to-end value prefix prediction (ablation) | EfficientZero · Atari 100k (26 games) | normed mean: 1.482 -> 1.943 | Helped |
| add end-to-end value prefix prediction (ablation) | EfficientZero · Atari 100k (26 games) | normed median: 0.552 -> 1.090 | Helped |
| add model-based off-policy correction (ablation) | EfficientZero · Atari 100k (26 games) | normed mean: 1.475 -> 1.943 | Helped |
| add model-based off-policy correction (ablation) | EfficientZero · Atari 100k (26 games) | normed median: 0.836 -> 1.090 | Helped |
| add off-policy correction to value targets (ablation) | EfficientZero · Atari UpNDown value error (current state) | L1 value error: 0.765 -> 0.533 | Helped |
| remove data augmentation in consistency component (ablation) | EfficientZero · 8 Atari games, 100k steps | score: Asterix 6218.8 -> 13884.0; Breakout 388.8 -> 365.2; Pong 19.8 -> 13.9 | Mixed |
| remove dynamic horizon from off-policy correction (ablation) | EfficientZero · Atari Qbert 100k | score: 15268.8 -> 7945 | Hurt |
| remove MCTS root value from off-policy correction (ablation) | EfficientZero · Atari Qbert 100k | score: 15268.8 -> 14738.0 | No clear effect |
| replace direct reward prediction with value prefix (ablation) | EfficientZero · Pong static dataset | validation error 5-step unroll | Helped |
Closest papers by meaning
- EfficientZero V2: Mastering Discrete and Continuous Control with Limited Data (ICML 2024 · similarity 0.91)
- Scaling Offline Model-Based RL via Jointly-Optimized World-Action Model Pretraining (ICLR 2025 · similarity 0.87)
- Investigating Pre-Training Objectives for Generalization in Vision-Based Reinforcement Learning (ICML 2024 · similarity 0.87)
- SRL: Scaling Distributed Reinforcement Learning to Over Ten Thousand Cores (ICLR 2024 · similarity 0.85)
- FASTER: Value-Guided Sampling for Fast RL (NeurIPS 2026 · similarity 0.85)
- The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning (NeurIPS 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.