Research map / Reinforcement learning / Visual and representation RL
Policy Information Capacity: Information-Theoretic Measure for Task Complexity in Deep Reinforcement Learning
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 POIC rather than reward marginal entropy as the solvability indicator | RL benchmark environments · OpenAI Gym, MuJoCo, and DM Control benchmarks | Pearson correlation with algorithm-based normalized score: -0.349 -> 0.807 | Helped |
| Use POIC rather than return variance as the solvability indicator | RL benchmark environments · OpenAI Gym, MuJoCo, and DM Control benchmarks | Pearson correlation with algorithm-based normalized score: 0.372 -> 0.807 | Helped |
| Use PIC rather than return variance as the solvability indicator | RL benchmark environments · OpenAI Gym, MuJoCo, and DM Control benchmarks | Pearson correlation with random-sampling-based normalized score: 0.314 -> 0.401 | Helped |
| Increase the synthetic MDP horizon from T=1 to T=3 | Synthetic MDP · Synthetic MDP | Normalized Score: 0.451 -> 0.112 | Hurt |
| Increase the synthetic MDP horizon from T=1 to T=3 | Synthetic MDP · Synthetic MDP | PIC: 0.087 -> 0.050 | Hurt |
| Increase the synthetic MDP horizon from T=1 to T=3 | Synthetic MDP · Synthetic MDP | POIC: 0.087 -> 0.049 | Hurt |
| Use a high-POIC initial policy-parameter prior (μ=0) rather than low-POIC priors | Evolution Strategies · Synthetic MDP, T=3 | Learning speed | Helped |
| Use a high-PIC initial policy-parameter prior (μ=0) rather than low-PIC priors | Evolution Strategies · Synthetic MDP, T=3 | Learning speed | Helped |
| Use PIC as the metric instead of marginal reward entropy | Policy-parameter priors · Synthetic MDP, T=3 | Pearson correlation with normalized score: R=0.95; R=0.78; marginal entropy R=0.99; R=-0.83 | Helped |
| Use POIC as the metric instead of marginal optimality entropy | Policy-parameter priors · Synthetic MDP, T=3 | Pearson correlation with normalized score: R=0.94; R=0.69; marginal entropy R=0.99; R=-0.82 | Helped |
| Choose reward shaping with smooth L1 or L2 distance rewards rather than Fraction or Sparse rewards | Reacher and Pointmaze · Reacher and Pointmaze | Robustness to reward hyper-parameters | Helped |
Closest papers by meaning
- Rethinking Model-based, Policy-based, and Value-based Reinforcement Learning via the Lens of Representation Complexity (NeurIPS 2024 · similarity 0.88)
- Representation Learning Enables Scalable Multitask Deep Reinforcement Learning (NeurIPS 2026 · similarity 0.86)
- Towards an Information Theoretic Framework of Context-Based Offline Meta-Reinforcement Learning (NeurIPS 2024 · similarity 0.86)
- Turning Sand to Gold: Recycling Data to Bridge On-Policy and Off-Policy Learning via Causal Bound (NeurIPS 2025 · similarity 0.86)
- DeepSearch: Overcome the Bottleneck of Reinforcement Learning with Verifiable Rewards via Monte Carlo Tree Search (ICLR 2026 · similarity 0.86)
- Not Only Where, But When: Temporal Scheduling for RLVR (NeurIPS 2026 · similarity 0.86)
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.