Research map / Reinforcement learning / Policy gradient and actor-critic
Generalized Weighted Path Consistency for Mastering Atari Games
Dengwei Zhao, Shikui Tu, Lei Xu
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 |
|---|---|---|---|
| add weighting to the path consistency target (G-PCZero -> GW-PCZero) (ablation) | PCZero · Atari 100k | mean human normalized score: 1.378 -> 1.980 | Improved |
| add path consistency (EfficientZero† -> G-PCZero) (ablation) | EfficientZero · Atari 100k | mean human normalized score: 1.218 -> 1.378 | Improved |
| make weighting coefficient ca adjustable (ablation) | GW-PCZero · Atari 100k | mean human normalized score: 1.980 -> 2.066 | Improved |
| restore 120k training steps and MCTS value correction at 4x MCTS cost | EfficientZero · Atari 100k | mean human normalized score: 1.218 -> 1.943 | Improved |
| set ca 0.0 -> 0.1 (ablation) | GW-PCZero · Boxing | game score: 11.8 -> 41.6 | Improved |
| set ca 0.1 -> 0.2 (ablation) | GW-PCZero · Boxing | game score: 41.6 -> 53.8 | Improved |
| set ca 0.1 -> 0.2 (ablation) | GW-PCZero · Krull | game score: 7782.0 -> 2182.5 | Hurt |
| set ca 0.1 -> 0.2 (ablation) | GW-PCZero · Qbert | game score: 13651.6 -> 11269.5 | Hurt |
| set ca 0.0 -> 0.1 (ablation) | GW-PCZero · Kangaroo | game score: 1012.5 -> 262.5 | Hurt |
| set ca 0.0 -> 0.05 (ablation) | GW-PCZero · Asterix | game score: 15712.5 -> 22131.3 | Improved |
| set ca 0.0 -> 0.1 (ablation) | GW-PCZero · Seaquest | game score: 838.1 -> 768.1 | Hurt |
| set ca 0.0 -> 0.2 (ablation) | GW-PCZero · Alien | game score: 761.3 -> 982.5 | Improved |
Closest papers by meaning
- Efficient Learning for AlphaZero via Path Consistency (ICML 2022 · similarity 0.90)
- Natural Value Approximators: Learning when to Trust Past Estimates (NeurIPS 2017 · similarity 0.85)
- Relative Entropy Pathwise Policy Optimization (ICLR 2026 · similarity 0.85)
- Successor Uncertainties: Exploration and Uncertainty in Temporal Difference Learning (NeurIPS 2019 · similarity 0.85)
- On Bonus Based Exploration Methods In The Arcade Learning Environment (ICLR 2020 · similarity 0.84)
- Towards Global Optimality for Practical Average Reward Reinforcement Learning without Mixing Time Oracles (ICML 2024 · similarity 0.84)
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.
