Research map / Probabilistic and scientific modeling / Treatment effects
On Inductive Biases for Heterogeneous Treatment Effect Estimation
Alicia Curth, Mihaela van der Schaar
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 |
|---|---|---|---|
| Encourage shared structure across potential-outcome functions using regularization | TNet and TARNet · Simulations A and B | CATE RMSE | Improved |
| Use additive reparametrization, expressing treatment outcome as control outcome plus treatment effect | TNet · Simulation A, especially with smaller treatment samples | CATE RMSE | Improved |
| Use additive reparametrization when treated outcome is simpler than control outcome | TNet · Simulation B | CATE RMSE | Hurt |
| Learn shared and private subspaces with FlexTENet | TARNet · Simulations A and B | CATE RMSE | Improved |
| Remove orthogonal regularization from FlexTENet and set lambda_o=0 (ablation) | FlexTENet · Simulations A and B | CATE RMSE | Hurt |
| Set lambda_1 < lambda_2, regularizing CATE more heavily than potential outcomes (ablation) | Soft regularization and FlexTENet · Setup A, n1=200 | CATE RMSE | Improved |
| Add orthogonal regularization between shared and private subspaces (ablation) | FlexTENet · Setup A, n1=200 | CATE RMSE | Improved |
| Combine orthogonal regularization with lambda_1 < lambda_2 (ablation) | FlexTENet · Setup A, n1=200 | CATE RMSE | Improved |
| Use the flexible architecture with smaller control sample size | FlexTENet · Setups A and B, smaller n0 | CATE RMSE | Mixed |
| Apply the soft regularizer to DragonNet | DragonNet · Setups A and B | CATE RMSE | Improved |
| Use FlexTENet rather than SNet, with comparable shared/private feature spaces | SNet · Setups A and B | CATE RMSE | Improved |
| Use reparametrization on IHDP setup C, where additive treatment effects and simpler CATE do not hold | TNet · IHDP setup C | normalized RMSE CATE: 0.32 (.01) -> 0.39 (.01) in-sample; 0.34 (.01) -> 0.40 (.01) out-of-sample | Hurt |
Closest papers by meaning
- Continuous Treatment Effect Estimation Using Gradient Interpolation and Kernel Smoothing (AAAI 2024 · similarity 0.89)
- ODE Discovery for Longitudinal Heterogeneous Treatment Effects Inference (ICLR 2024 · similarity 0.88)
- Bounds on Representation-Induced Confounding Bias for Treatment Effect Estimation (ICLR 2024 · similarity 0.88)
- A Non-parametric Direct Learning Approach to Heterogeneous Treatment Effect Estimation under Unmeasured Confounding (NeurIPS 2024 · similarity 0.88)
- PairNet: Training with Observed Pairs to Estimate Individual Treatment Effect (ICML 2024 · similarity 0.88)
- Learning Representations of Instruments for Partial Identification of Treatment Effects (ICML 2025 · 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.
