Research map / Reinforcement learning / Policy gradient and actor-critic
Max-Min Off-Policy Actor-Critic Method Focusing on Worst-Case Robustness to Model Misspecification
Takumi Tanabe, Rei Sato, Kazuto Fukuchi, Jun Sakuma, Youhei Akimoto
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 N=1 uncertainty parameter instead of N=5 (ablation) | M2TD3 · InvertedPendulum 1 | worst-case return (x10^2): 8.22 -> 5.77 | Hurt |
| increase N from 5 to 10 (ablation) | M2TD3 · InvertedPendulum 1 | worst-case return (x10^2): 8.22 -> 9.07 | Improved |
| increase N from 5 to 10 (ablation) | M2TD3 · InvertedPendulum 1 | average return (x10^2): 9.66 -> 9.89 | Improved |
| remove distance-based refreshing strategy (w/o DRS) (ablation) | M2TD3 · InvertedPendulum 1 | worst-case return (x10^2): 8.22 -> 9.07 | No clear effect |
| remove distance-based refreshing strategy (w/o DRS) (ablation) | M2TD3 · InvertedPendulum 1 | average return (x10^2): 9.66 -> 9.08 | Hurt |
| remove probability-based refreshing strategy (w/o PRS) (ablation) | M2TD3 · InvertedPendulum 1 | worst-case return (x10^2): 8.22 -> 5.41 | Hurt |
| remove probability-based refreshing strategy (w/o PRS) (ablation) | M2TD3 · InvertedPendulum 1 | average return (x10^2): 9.66 -> 6.90 | Hurt |
| use M2TD3 instead of DR (TD3) for worst-case | M2TD3 · InvertedPendulum 1 | worst-case return (x10^2): 3.18 -> 8.22 | Improved |
| use M2TD3 instead of DR (TD3) | M2TD3 · Ant 2 | worst-case return (x10^3): 1.64 -> 4.13 | Improved |
| use M2TD3 instead of DR (TD3) | M2TD3 · Small HalfCheetah 1 | worst-case return (x10^3): 6.76 -> 5.27 | Hurt |
| use M2TD3 instead of DR (TD3) | M2TD3 · HumanoidStandup 1 | worst-case return (x10^4): 9.68 -> 9.33 | Hurt |
| use M2TD3 instead of DR (TD3) | M2TD3 · Small Hopper 1 | worst-case return (x10^3): 3.42 -> 2.88 | Hurt |
Closest papers by meaning
- Policy Gradient Method For Robust Reinforcement Learning (ICML 2022 · similarity 0.89)
- Model-Based Offline Reinforcement Learning with Local Misspecification (AAAI 2023 · similarity 0.88)
- A Unified Principle of Pessimism for Offline Reinforcement Learning under Model Mismatch (NeurIPS 2024 · similarity 0.87)
- Minimax Optimal and Computationally Efficient Algorithms for Distributionally Robust Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Deterministic Uncertainty Propagation for Improved Model-Based Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Constrained Reinforcement Learning Under Model Mismatch (ICML 2024 · similarity 0.87)
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.
