Research map / Reinforcement learning / Policy gradient and actor-critic
Personalized Policy Learning through Discrete Experimentation
Zhiqi Zhang, Zhiyu Zeng, Ruohan Zhan, Dennis Zhang
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 |
|---|---|---|---|
| use DLPT instead of pure deep learning (PDL) for ATE extrapolation | structured DNN · Held-out arm ATE recovery | MAPE (%): 7.18 -> 2.25 | Improved |
| use DLPT instead of linear regression for ATE extrapolation | structured DNN · Held-out arm ATE recovery | MAPE (%): 38.23 -> 2.25 | Improved |
| use DLPT instead of GRF T-learner for ATE extrapolation | structured DNN · Held-out arm ATE recovery | MAPE (%): 40.43 -> 2.25 | Improved |
| use DLPT instead of PDL for ATE extrapolation | structured DNN · Held-out arm ATE recovery | MSE: 32.82 -> 3.46 | Improved |
| use DLPT instead of PDL for policy learning | structured DNN · Policy learning, finite discrete class | MPR (%): 6.32 -> 0.48 | Improved |
| use DLPT instead of best benchmark for optimal-arm accuracy | structured DNN · Policy learning, finite discrete class | Accuracy (%): 25.50 -> 73.08 | Improved |
| use DLPT instead of best benchmark for weighted accuracy | structured DNN · Policy learning, finite discrete class | Weighted accuracy (%): 33.79 -> 79.83 | Improved |
| use DLPT instead of PDL in finite observed policy class | structured DNN · Semi-synthetic, Finite Observed policy class | MPR (%): 27.58 -> 17.96 | Improved |
| use DLPT instead of PDL in DNN policy class | structured DNN · Semi-synthetic, DNN policy class | MPR (%): 7.42 -> 0.99 | Improved |
| use continuous DNN policy class instead of finite observed policy class for DLPT | DLPT · Semi-synthetic | MPR (%): 17.96 -> 0.99 | Improved |
| use DNN policy class instead of finite observed class for simple parametric estimators (LR, LogR, LCM) | LR/LogR/LCM · Semi-synthetic | MPR (%): 29.89 -> 12.65 | Hurt |
| use linear policy class instead of finite observed class for DLPT | DLPT · Semi-synthetic | MPR (%): 17.96 -> 0.01 | Improved |
Closest papers by meaning
- Personalized Reinforcement Learning with a Budget of Policies (AAAI 2024 · similarity 0.86)
- Test-Time Personalization: A Diagnostic Framework and Probabilistic Fix for Scaling Failures (NeurIPS 2026 · similarity 0.85)
- Towards Safe Policy Learning under Partial Identifiability: A Causal Approach (AAAI 2024 · similarity 0.84)
- Policy Filtration for RLHF to Mitigate Noise in Reward Models (ICML 2025 · similarity 0.83)
- Learning Personalized Decision Support Policies (AAAI 2025 · similarity 0.83)
- Improving Reward Model Generalization from Adversarial Process Enhanced Preferences (ICML 2025 · similarity 0.83)
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.
