Research map / Probabilistic and scientific modeling / Treatment effects
Efficient augmentation and relaxation learning for individualized treatment rules using observational data
Ying-Qi Zhao, Eric B. Laber, Yang Ning, Sumona Saha, Bruce E. Sands
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 |
|---|---|---|---|
| add outcome regression augmentation (EARL) instead of IPWE-only outcome weighted learning | EARL · Simulation (Scenarios 1-3) | value of estimated rule | Improved |
| use convex relaxation (EARL) instead of direct AIPWE maximization via genetic algorithm | AIPWE value search · Simulation | bias, variance and runtime: approximately 200 times slower | Improved |
| correctly specify the Q-function model in Q-learning (ablation) | Q-learning · Simulation | value of estimated rule | Mixed |
| misspecify the Q-function model rather than the propensity model (ablation) | EARL and AIPWE · Simulation | value of estimated rule | Hurt |
| bound propensity scores away from zero (Scenario 2 versus Scenario 1) (ablation) | EARL · Simulation | value of estimated rule | Improved |
| set propensity to constant 0.025 with correct propensity but wrong outcome model (CI) versus the reverse (IC) (ablation) | EARL · Simulation Scenario 3 | value of estimated rule: CI outperformed IC, especially at small sample size | Improved |
| use nonparametric working models for propensity score instead of parametric (ablation) | EARL · Simulation (Web Appendix F) | value of estimated rule | Improved |
| use hinge loss instead of smooth surrogate losses when the regression model is incorrect (ablation) | EARL · Simulation (Web Appendix F) | value of estimated rule | Improved |
| use EARL instead of observed treatment allocation | EARL logistic · OSCCAR IBD registry | estimated disease activity score: 2.24 -> 1.75 | Improved |
| use EARL logistic instead of Q-learning | EARL logistic · OSCCAR IBD registry | estimated disease activity score: 1.80 -> 1.75 | Improved |
| use EARL logistic instead of OWL logistic | EARL logistic · OSCCAR IBD registry | estimated disease activity score and patients on top-down: 1.75 vs 1.75; 161 vs 225 patients recommended top-down | No clear effect |
| switch EARL surrogate loss among logistic, hinge, squared hinge, exponential (ablation) | EARL · OSCCAR IBD registry | rule agreement: agreement 0.996 between logit and hinge/sqhinge; 0.588 with exponential | No clear effect |
Closest papers by meaning
- An effective framework for estimating individualized treatment rules (NeurIPS 2024 · similarity 0.90)
- Continuous Treatment Effect Estimation Using Gradient Interpolation and Kernel Smoothing (AAAI 2024 · similarity 0.86)
- Doubly Robust Fusion of Many Treatments for Policy Learning (ICML 2025 · similarity 0.86)
- Meta-Learners for Partially-Identified Treatment Effects Across Multiple Environments (ICML 2024 · similarity 0.85)
- Quantifying Aleatoric Uncertainty of the Treatment Effect: A Novel Orthogonal Learner (NeurIPS 2024 · similarity 0.85)
- Learning Representations of Instruments for Partial Identification of Treatment Effects (ICML 2025 · similarity 0.85)
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.
