Research map / Probabilistic and scientific modeling / Treatment effects
Deep Jump Learning for Off-Policy Evaluation in Continuous Treatment Settings
Hengrui Cai, Chengchun Shi, Rui Song, Wenbin Lu
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 adaptive discretization of treatment space with DJL instead of a single-bandwidth kernel | DR value estimator · Toy example V2 (pi(X)>0.5) | absolute error: 1.09 -> 0.09 | Improved |
| Use adaptive discretization instead of a single large bandwidth h=1 | value estimator · Toy example V1 | absolute error: 0.40 -> 0.31 | Improved |
| Use small bandwidth h=0.4 versus h=1 in kernel estimator | kernel-based OPE · Toy example V1 and V2 | absolute error: 0.40 -> 0.50 | Mixed |
| Replace kernel baselines with DJL | policy value estimator · Warfarin calibrated data | mean squared error: 0.943 -> 0.240 | Improved |
| Replace SLOPE with DJL | policy value estimator · Warfarin calibrated data | absolute error: 0.611 -> 0.259 | Improved |
| Increase sample size from 50 to 300 | DJL · Scenario 1 | absolute error: 0.445 -> 0.209 | Improved |
| Increase sample size from 50 to 300 | DJL · Scenario 3 | absolute error: 2.014 -> 1.267 | Improved |
| Use kernel baseline 1 instead of DJL | kernel OPE · Scenario 3, n=300 | absolute error: 1.267 -> 1.151 | Improved |
| Compare DJL with SLOPE at small sample size | DJL · Scenario 1, n=50 | absolute error: 0.445 -> 0.392 | No clear effect |
| Replace DJL with SLOPE | policy value estimator · Scenario 2, n=300 | absolute error: 0.411 -> 1.209 | Hurt |
| Choose number of initial intervals m=n/10 (ablation) | DJL · Scenario 1 | absolute error and computational cost | Improved |
| Replace first kernel baseline with DJL | DJL · Scenario 1, n=50 vs DJL n=300 | minutes: 365 -> 14 | Improved |
Closest papers by meaning
- Reliable Off-Policy Learning for Dosage Combinations (NeurIPS 2023 · similarity 0.87)
- Continuous Treatment Effect Estimation Using Gradient Interpolation and Kernel Smoothing (AAAI 2024 · similarity 0.85)
- Enhancing Treatment Effect Estimation via Active Learning: A Counterfactual Covering Perspective (ICML 2025 · similarity 0.84)
- EpiCare: A Reinforcement Learning Benchmark for Dynamic Treatment Regimes (NeurIPS 2024 · similarity 0.83)
- Adversarially Balanced Representation for Continuous Treatment Effect Estimation (AAAI 2024 · similarity 0.83)
- Longitudinal Targeted Minimum Loss-based Estimation with Temporal-Difference Heterogeneous Transformer (ICML 2024 · 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.
