Research map / Reinforcement learning / Policy gradient and actor-critic
Empirical Study of Off-Policy Policy Evaluation for 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 MAGIC with FQE rather than the other evaluated estimators | OPE estimators · All experiments | Near-top Frequency: 0.300211 | Helped |
| Increase horizon from T = 10 to T = 100 at the same mild policy mismatch | Graph OPE · Graph, N = 50 | Relative MSE: 1.8E-3 -> 6.0E-2 | Hurt |
| Increase policy mismatch while keeping T = 10 | Graph OPE · Graph, N = 50 | Relative MSE: 1.8E-3 -> 6.6E-1 | Hurt |
| Use IH instead of IS under increased horizon and mild policy mismatch | Graph OPE · Graph, T = 100, N = 50 | Relative MSE: 1.7E-2 -> 4.7E-4 | Helped |
| Use IH instead of PDIS under large policy mismatch | Graph OPE · Graph, T = 10, N = 50 | Relative MSE: 5.4E-1 -> 1.4E-2 | Helped |
| Use weighted importance sampling rather than standard importance sampling | IPS estimators · Graph, T = 100, N = 50, mild policy mismatch | Relative MSE: 1.7E-2 -> 9.5E-4 | Helped |
| Use FQE as the value estimator in MAGIC rather than direct FQE alone | OPE estimators · All experiments | Near-top Frequency: 0.236786 -> 0.300211 | Helped |
| Use a hybrid estimator instead of a direct method under severe representation mismatch | OPE estimators · Graph-POMDP | Near-top Frequency | Helped |
| Use hybrid methods instead of direct methods when behavior policy estimates are poor and data are limited | OPE estimators · Pixel-Gridworld | Relative MSE | Hurt |
| Use hybrid methods instead of direct methods in stochastic environments with limited data | OPE estimators · Stochastic Graph | Relative MSE | Hurt |
| Use weighted doubly robust rather than standard doubly robust | Hybrid estimators · Across benchmark conditions | Relative MSE | Helped |
| Use direct methods in Enduro's high-dimensional, long-horizon, sparse-reward setting | OPE estimators · Enduro | Relative MSE | No clear effect |
Closest papers by meaning
- Model Selection for Off-policy Evaluation: New Algorithms and Experimental Protocol (NeurIPS 2025 · similarity 0.91)
- Demystifying the Paradox of Importance Sampling with an Estimated History-Dependent Behavior Policy in Off-Policy Evaluation (ICML 2025 · similarity 0.90)
- Statistical Tractability of Off-policy Evaluation of History-dependent Policies in POMDPs (ICLR 2025 · similarity 0.90)
- Abstract Reward Processes: Leveraging State Abstraction for Consistent Off-Policy Evaluation (NeurIPS 2024 · similarity 0.89)
- Efficient Policy Evaluation with Offline Data Informed Behavior Policy Design (ICML 2024 · similarity 0.89)
- Kernel Metric Learning for In-Sample Off-Policy Evaluation of Deterministic RL Policies (ICLR 2024 · similarity 0.88)
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.