Research map / Reinforcement learning / Policy gradient and actor-critic
Using Options and Covariance Testing for Long Horizon Off-Policy Policy Evaluation
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 options of length n instead of primitive actions | PDIS · NoisyTaxi | MSE: an order of magnitude better | Helped |
| Use 3-step options rather than shorter options (ablation) | PDIS · NoisyTaxi | MSE | Helped |
| Partition trajectories at the stationary pickup option and estimate each segment separately | PDIS · NoisyTaxi | MSE: an order of an order of magnitude reduction | Helped |
| Drop all but the most recent weights, selecting the retained window with the covariance test | PDIS (INCRIS) · Simple MDP with dependent sub-episodes | MSE: an order of magnitude lower | Helped |
| Drop all but the most recent weights, selecting the retained window with the covariance test | PDIS (INCRIS) · Simple MDP with dependent sub-episodes | MSE: two orders of magnitude improvement | Helped |
| Use options with unchanged underlying policies and minimum duration J | PDIS · Long-horizon OPE | worst-case variance growth: Ω(2^H) to Ω(2^(H/J)) | Helped |
| Keep weights only for options whose underlying policies changed | PDIS · Long-horizon OPE | worst-case variance growth: Ω(2^H) to Ω(2^K) | Helped |
| Drop weights before a stationary option when estimating rewards after it | PDIS · Options-based OPE | bias | Helped |
Closest papers by meaning
- State Relevance for Off-Policy Evaluation (ICML 2021 · similarity 0.92)
- Understanding the Curse of Horizon in Off-Policy Evaluation via Conditional Importance Sampling (ICML 2020 · similarity 0.92)
- Demystifying the Paradox of Importance Sampling with an Estimated History-Dependent Behavior Policy in Off-Policy Evaluation (ICML 2025 · similarity 0.88)
- Towards Assessing and Benchmarking Risk-Return Tradeoff of Off-Policy Evaluation (ICLR 2024 · similarity 0.87)
- Off-policy Evaluation Beyond Overlap: Sharp Partial Identification Under Smoothness (ICML 2024 · similarity 0.87)
- OPERA: Automatic Offline Policy Evaluation with Re-weighted Aggregates of Multiple Estimators (NeurIPS 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.