Research map / Probabilistic and scientific modeling / Treatment effects
Learning Counterfactual Representations for Estimating Individual Dose-Response Curves
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 DRNet's hierarchical dosage-stratified architecture instead of a flat MLP (ablation) | Dose-response estimation · News-2, News-4, News-8, News-16, MVICU, TCGA | √MISE: 15.3 ± 0.1 -> 8.0 ± 0.1 (News-2); 14.5 ± 0.0 -> 11.6 ± 0.1 (News-4); 13.9 ± 0.1 -> 10.2 ± 0.1 (News-8); 14.0 ± 0.0 -> 10 | Helped |
| Use dosage-stratified DRNet heads instead of TARNET's treatment-specific heads with dosage input (ablation) | Dose-response estimation · News-2, News-4, News-8, News-16, MVICU, TCGA | √MISE: 15.5 ± 0.1 -> 8.0 ± 0.1 (News-2); 15.4 ± 0.0 -> 11.6 ± 0.1 (News-4); 14.7 ± 0.1 -> 10.2 ± 0.1 (News-8); 14.7 ± 0.1 -> 10 | Helped |
| Add Wasserstein regularization between treatment groups (ablation) | DRNet · News-2, News-4, News-8, News-16 | √MISE: 11.6 ± 0.1 -> 11.5 ± 0.0 (News-4); 10.2 ± 0.1 -> 10.0 ± 0.0 (News-8); 10.3 ± 0.0 -> 10.2 ± 0.0 (News-16) | Helped |
| Add Wasserstein regularization between treatment groups (ablation) | DRNet · MVICU and TCGA | √MISE: 31.1 ± 0.4 -> 32.9 ± 2.9 (MVICU); 9.6 ± 0.0 -> 10.2 ± 0.9 (TCGA) | Hurt |
| Add propensity dropout (ablation) | DRNet · News-4, News-8, News-16, MVICU, TCGA | √MISE: 11.6 ± 0.1 -> 12.2 ± 0.1 (News-4); 10.2 ± 0.1 -> 10.6 ± 0.2 (News-8); 10.3 ± 0.0 -> 10.3 ± 0.1 (News-16); 31.1 ± 0.4 -> | Hurt |
| Add batch-level matching (ablation) | DRNet · News-4, News-8, News-16, MVICU, TCGA | √MISE: 11.6 ± 0.1 -> 12.2 ± 0.1 (News-4); 10.2 ± 0.1 -> 11.4 ± 0.3 (News-8); 10.3 ± 0.0 -> 12.3 ± 0.3 (News-16); 31.1 ± 0.4 -> | Mixed |
| Add whole-training-set matching preprocessing (ablation) | DRNet · News-4, News-8, News-16, MVICU, TCGA | √MISE: 11.6 ± 0.1 -> 12.2 ± 0.2 (News-4); 10.2 ± 0.1 -> 11.5 ± 0.2 (News-8); 10.3 ± 0.0 -> 12.2 ± 0.3 (News-16); 31.1 ± 0.4 -> | Hurt |
| Add batch-level matching rather than no additional regularization (ablation) | DRNet · News-2 | √MISE | Helped |
| Do not append dosage repeatedly in the head layers (ablation) | DRNet · News-4 and News-8 | √MISE: 11.6 ± 0.1 -> 11.9 ± 0.2 (News-4); 10.2 ± 0.1 -> 10.3 ± 0.1 (News-8) | Hurt |
| Do not append dosage repeatedly in the head layers (ablation) | DRNet · News-16 and MVICU | √MISE: 10.3 ± 0.0 -> 10.4 ± 0.1 (News-16); 31.1 ± 0.4 -> 31.0 ± 0.3 (MVICU) | Mixed |
| Increase the number of dosage strata (ablation) | DRNet · MVICU | MISE | Helped |
| Increase the number of dosage strata (ablation) | DRNet · MVICU | training and prediction time | Hurt |
Closest papers by meaning
- Contrastive Balancing Representation Learning for Heterogeneous Dose-Response Curves Estimation (AAAI 2024 · similarity 0.91)
- Continuous Treatment Effect Estimation Using Gradient Interpolation and Kernel Smoothing (AAAI 2024 · similarity 0.88)
- Adversarially Balanced Representation for Continuous Treatment Effect Estimation (AAAI 2024 · similarity 0.87)
- Counterfactual Contrastive Learning with Normalizing Flows for Robust Treatment Effect Estimation (ICML 2025 · similarity 0.86)
- Enhancing Treatment Effect Estimation via Active Learning: A Counterfactual Covering Perspective (ICML 2025 · similarity 0.85)
- Predicting Dose-Response Curves with Deep Neural Networks (ICML 2024 · 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.