Research map / Learning theory and algorithms / Prediction and decision losses
Stage-Aware Learning for Dynamic Treatments
Hanwen Ye, Wenzhuo Zhou, Ruoqing Zhu, Annie Qu
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 stage-aware partial matching (SAL) instead of IPWE full-matching (BOWL) | SAL · Simulation T=10, n=5000, full-matching probability 0.001 | estimated total reward: 0.347 -> 0.624 | Improved |
| Use SAL instead of BOWL at high full-matching probability | SAL · Simulation T=10, n=5000, full-matching probability 0.100 | estimated total reward: 0.659 -> 0.686 | Improved |
| Decrease sample size from 5000 to 500 (SAL margin over best competitor) | SAL · Simulation nonlinear regime, T=5 | estimated total reward: 10.081% -> 27.924% | Improved |
| Increase number of stages from 5 to 10 (SAL margin over best competitor) | SAL · Simulation nonlinear regime, n=5000 | estimated total reward: 10.081% -> 20.541% | Improved |
| Add stage importance scores to SAL (SWL vs SAL) (ablation) | SWL · Simulation nonlinear regime, no important stage | estimated total reward: 0.673 -> 0.671 | No clear effect |
| Add stage importance scores to SAL when few stages are important (ablation) | SWL · Simulation linear homogeneous, n=500, T=10 | estimated total reward | Improved |
| Use SWL instead of SAL on ICU 8-stage data (ablation) | SWL · UC CORDS ICU, 8 stages | hospitalization days: 4.913 -> 6.749 | Hurt |
| Use SWL instead of SAL on inpatient 8-stage data (ablation) | SWL · UC CORDS Inpatient, 8 stages | hospitalization days: 6.704 -> 5.664 | Improved |
| Use SWL instead of SAL on ICU 5-stage data (ablation) | SWL · UC CORDS ICU, 5 stages | hospitalization days: 7.175 -> 5.855 | Improved |
| Use SWL instead of BOWL on inpatient 10-stage data | SWL · UC CORDS Inpatient, 10 stages | hospitalization days: 15.129 -> 4.817 | Improved |
| Linear homogeneous regime: SWL vs BOWL matching accuracy at n=1000 | SWL · Simulation linear homogeneous, T=5 | matching accuracy: 74.71% -> 97.36% | Improved |
| Use nonlinear heterogeneous instead of linear homogeneous regime (ablation) | SAL/SWL · Simulation, T=5 | estimated total reward | Hurt |
Closest papers by meaning
- Evaluating and Learning Optimal Dynamic Treatment Regimes under Truncation by Death (NeurIPS 2025 · similarity 0.86)
- Position: Reinforcement Learning in Dynamic Treatment Regimes Needs Critical Reexamination (ICML 2024 · similarity 0.85)
- SAFER: A Calibrated Risk-Aware Multimodal Recommendation Model for Dynamic Treatment Regimes (ICML 2025 · similarity 0.85)
- Doubly Robust Fusion of Many Treatments for Policy Learning (ICML 2025 · similarity 0.84)
- An effective framework for estimating individualized treatment rules (NeurIPS 2024 · similarity 0.84)
- Model-agnostic meta-learners for estimating heterogeneous treatment effects over time (ICLR 2025 · similarity 0.84)
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.
