Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Temporal-Spatial Causal Interpretations for Vision-Based Reinforcement Learning
Wenjie Shi, Gao Ming Huang, Shiji Song, Cheng Wu
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 TSCI-discovered features instead of SARFA features as the only pixels the policy sees | policy trained on masked pixels · Enduro | average return: 2369 -> 2903 | Improved |
| use TSCI features instead of perturbation-based features | policy trained on masked pixels · Enduro | average return: 1741 -> 2903 | Improved |
| use TSCI features instead of gradient-based features | policy trained on masked pixels · Enduro | average return: 819 -> 2903 | Improved |
| use TSCI features instead of SARFA features | policy trained on masked pixels · Seaquest | average return: 2085 -> 2517 | Improved |
| use TSCI features instead of perturbation-based features | policy trained on masked pixels · Seaquest | average return: 1830 -> 2517 | Improved |
| use TSCI features instead of gradient-based features | policy trained on masked pixels · Seaquest | average return: 836 -> 2517 | Improved |
| use temporal causal objective instead of CXPlain or imitation objectives (ablation) | TSCI causal discovery network · Atari behavior-matching reliability | behavior matching errors and temporal causality error | Improved |
| use temporal causal objective instead of CXPlain or imitation objectives (ablation) | TSCI causal discovery network · Atari | mean normalized return | No clear effect |
| mask semantic information in more of the stacked frames (4 -> 34 -> 234 -> 1234) (ablation) | PPO recurrent agent · Atari counterfactual intervention | average return | Hurt |
| mask semantic information in a single previous frame (schemes 1, 2, 3) (ablation) | PPO recurrent agent · Atari counterfactual intervention | average return | No clear effect |
| stack only the current frame (no frame stacking) in a recurrent agent (ablation) | recurrent PPO agent · Atari | average return | Hurt |
| stack more consecutive frames in recurrent agent input (ablation) | recurrent PPO agent · Atari | average return | Improved |
Closest papers by meaning
- ERCI: An Explainable Experience Replay Approach with Causal Inference for Deep Reinforcement Learning (AAAI 2025 · similarity 0.85)
- Sim-to-Real Causal Transfer: A Metric Learning Approach to Causally-Aware Interaction Representations (CVPR 2025 · similarity 0.85)
- NoisyRollout: Reinforcing Visual Reasoning with Data Augmentation (NeurIPS 2025 · similarity 0.85)
- CaRiNG: Learning Temporal Causal Representation under Non-Invertible Generation Process (ICML 2024 · similarity 0.84)
- Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach (CVPR 2025 · similarity 0.84)
- DrM: Mastering Visual Reinforcement Learning through Dormant Ratio Minimization (ICLR 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.
