Research map / Reinforcement learning / Policy gradient and actor-critic
Model Selection in Batch Policy Optimization
Jonathan Lee, George Tucker, Ofir Nachum, Bo Dai
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 |
|---|---|---|---|
| Select actions pessimistically across all model classes | Complexity-Coverage Selection · Synthetic fully realizable contextual bandit | regret | Improved |
| Use the model class with the smallest hidden dimension | Single-class base algorithms · Synthetic fully realizable contextual bandit | regret | Improved |
| Select among model classes using SLOPE | SLOPE Method · Synthetic approximation error-complexity trade-off | regret | Improved |
| Select among model classes using hold-out regression loss | Hold-out Method · Synthetic approximation error-complexity trade-off | regret | Improved |
| Use hold-out selection instead of SLOPE | Model selection for approximation error-complexity trade-off · Synthetic approximation error-complexity trade-off, small data | regret | Improved |
| Balance approximation error, complexity, and coverage simultaneously | Batch policy optimization model selection · Contextual bandits with nested linear model classes | oracle-inequality competitiveness | Hurt |
| Ignore approximation error and select pessimistically across classes | Complexity-Coverage Selection · Contextual bandits | regret | Improved |
| Ignore comparator-specific coverage and use nested classes with SLOPE value estimates | SLOPE Method · Contextual bandits | regret | Improved |
| Ignore comparator-specific coverage and select by hold-out regression loss | Hold-out Method · Contextual bandits | regret | Improved |
Closest papers by meaning
- On the Optimality of Batch Policy Optimization Algorithms (ICML 2021 · similarity 0.91)
- Model Selection for Contextual Bandits (NeurIPS 2019 · similarity 0.89)
- Universal and data-adaptive algorithms for model selection in linear contextual bandits (ICML 2022 · similarity 0.89)
- Sample-Efficiency in Multi-Batch Reinforcement Learning: The Need for Dimension-Dependent Adaptivity (ICLR 2024 · similarity 0.87)
- Cost-Aware Near-Optimal Policy Learning (AAAI 2025 · similarity 0.87)
- Logarithmic Smoothing for Pessimistic Off-Policy Evaluation, Selection and Learning (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.
