Research map / Reinforcement learning / Visual and representation RL
Causal Dynamics Learning for Task-Independent State Abstraction
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 |
|---|---|---|---|
| Use CMI-based causal dependency inference instead of learned regularization masks | CDL · Chemical and Manipulation | causal graph accuracy: 100.0 ± 0.0 vs 99.4 ± 0.4; 100.0 ± 0.1 vs 99.7 ± 0.1; 99.1 ± 0.1 vs 97.7 ± 0.4; 90.2 ± 0.3 vs 84.4 ± 0.5 | Helped |
| Use PredDiff transition collection rather than a uniform policy | CDL · Manipulation | causal graph accuracy: 89.1 ± 0.6 -> 90.2 ± 0.3 | Helped |
| Use PredDiff transition collection rather than curiosity rewards | CDL · Manipulation | causal graph accuracy: 88.9 ± 0.2 -> 90.2 ± 0.3 | Helped |
| Use globally inferred causal dependencies rather than local parent selection | CDL · Chemical collider, OOD states | prediction accuracy: 35.7 -> 97.3 | Helped |
| Use causal dynamics and the derived state abstraction rather than dense dynamics with the full state | CDL · Unseen states in Chemical and Manipulation | prediction performance | Helped |
| Derive the state abstraction from the causal graph and omit action-irrelevant variables | CDL downstream task learning · Lift and Stack (Manipulation); Match (Chemical) | sample efficiency | Helped |
| Use CDL's causal state abstraction rather than dense models for policy learning | Downstream policies · OOD states in downstream tasks | episode reward | Helped |
Closest papers by meaning
- Building Minimal and Reusable Causal State Abstractions for Reinforcement Learning (AAAI 2024 · similarity 0.92)
- Fine-Grained Causal Dynamics Learning with Quantization for Improving Robustness in Reinforcement Learning (ICML 2024 · similarity 0.89)
- Invariant Causal Prediction for Block MDPs (ICML 2020 · similarity 0.89)
- Learning Causal Dynamics Models in Object-Oriented Environments (ICML 2024 · similarity 0.89)
- Learning Nonlinear Causal Reductions to Explain Reinforcement Learning Policies (ICLR 2026 · similarity 0.88)
- Reward-Consistent Dynamics Models are Strongly Generalizable for Offline Reinforcement Learning (ICLR 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.