Research map / Multi-agent systems and games / Human-AI and goal-conditioned
StateMask: Explaining Deep Reinforcement Learning through State Mask
Zelei Cheng, Xian Wu, Jiahao Yu, Wenhai Sun, Wenbo Guo, Xinyu Xing
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 StateMask instead of EDGE to guide attacks | adversarial attack · Pong | winning rate change: -63.04 -> -84.90 | Improved |
| use StateMask instead of Value-max to guide attacks | adversarial attack · Pong | winning rate change: -0.40 -> -84.90 | Improved |
| use StateMask instead of EDGE to guide attacks | adversarial attack · DouDizhu | winning rate change: -0.25 -> -29.94 | Improved |
| use StateMask instead of EDGE to guide attacks | adversarial attack · You-Shall-Not-Pass | winning rate change: -2.64 -> -35.93 | Improved |
| use StateMask instead of LazyMDP to guide attacks | adversarial attack · StarCraft II | winning rate change: -72.00 -> -83.00 | Improved |
| use StateMask instead of EDGE to guide patching | agent patching · You-Shall-Not-Pass | winning rate change: +0.07 -> +2.42 | Improved |
| use StateMask instead of EDGE to guide patching | agent patching · Connect 4 | winning rate change: +2.00 -> +2.40 | Mixed |
| use StateMask instead of LazyMDP to guide patching | agent patching · DouDizhu | winning rate change: +0.20 -> +0.47 | Improved |
| replace StateMask with EDGE on long trajectories | fidelity · DouDizhu and StarCraft II | fidelity score | Hurt |
| replace StateMask with value-based methods | fidelity · DouDizhu and StarCraft II | fidelity score | Hurt |
| vary sparsity weight w over {0, 1e-5, 1e-4, 1e-3, 1e-2} (ablation) | StateMask · Pong, You-Shall-Not-Pass | fidelity score | No clear effect |
| add sparsity constraint (w>0) (ablation) | StateMask · Pong, You-Shall-Not-Pass | mask ratio | Improved |
Closest papers by meaning
- RICE: Breaking Through the Training Bottlenecks of Reinforcement Learning with Explanation (ICML 2024 · similarity 0.88)
- ORCAID: Oblique Rule-Based Continuous-Action Interpretation for Deep RL Policies (NeurIPS 2026 · similarity 0.87)
- CrystalBox: Future-Based Explanations for Input-Driven Deep RL Systems (AAAI 2024 · similarity 0.86)
- Pure-Past Action Masking (AAAI 2024 · similarity 0.86)
- Hide to Guide: Learning via Semantic Masking (NeurIPS 2026 · similarity 0.85)
- Excluding the Irrelevant: Focusing Reinforcement Learning through Continuous Action Masking (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.
