Research map / Probabilistic and scientific modeling / Treatment effects
DSCF: Dual-Source Counterfactual Fusion for High-Dimensional Combinatorial Interventions
Jitong Dou, Lingrui Luo, Bing Zhu, Hengliang Luo, Mingjun Zhong, Yurong Cheng
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 |
|---|---|---|---|
| train on proxy counterfactual data Dpcf instead of observational Dobs (MLP) (ablation) | MLP · Synthetic ablation (gamma=1.0) | RMSE: 8.622 -> 7.636 | Improved |
| train on Dpcf instead of Dobs (MMoE) (ablation) | MMoE · Synthetic ablation (gamma=0.1) | RMSE: 2.988 -> 2.782 | Improved |
| add joint training on both sources (pcf head output) (ablation) | MMoE on Dpcf · Synthetic ablation (gamma=0.1) | RMSE: 2.782 -> 2.610 | Improved |
| add domain-guided fusion to joint MMoE (vs obs head output) (ablation) | joint MMoE · Synthetic ablation (gamma=1.0) | RMSE: 6.462 -> 6.223 | Improved |
| add domain-guided fusion to joint MMoE (vs averaging heads) (ablation) | joint MMoE · Synthetic ablation (gamma=0.1) | RMSE: 2.602 -> 2.603 | No clear effect |
| halve hidden dimensions (MMoE-lite) (ablation) | DSCF · Synthetic ablation (gamma=0.1) | RMSE: 2.603 -> 2.664 | Hurt |
| train jointly instead of fusing separately trained S-Learner and NNpcf (DSCF-Sep) | DSCF · Semi-synthetic (n=8M) | RMSE: 0.504 -> 0.353 | Improved |
| use DSCF instead of NNpcf | NNpcf · Semi-synthetic (n=0.1M) | RMSE: 0.870 -> 0.668 | Improved |
| use DSCF instead of NNpcf | NNpcf · Semi-synthetic (n=8M) | MAE: 0.288 -> 0.149 | Improved |
| use DSCF instead of H-Learner | H-Learner · Synthetic (p=30, gamma=0.1) | RMSE: 0.820 -> 0.983 | Hurt |
| use DSCF instead of H-Learner | H-Learner · Synthetic (p=20, gamma=1.0) | RMSE: 41.361 -> 6.640 | Improved |
| use reweighting (PW/VSR) or RMNet instead of matched proxy data (NNpcf) | NNpcf · Synthetic (p=10, gamma=0.1) | RMSE: 3.776 -> 4.339 (PW) | Hurt |
Closest papers by meaning
- An Empirical Examination of Balancing Strategy for Counterfactual Estimation on Time Series (ICML 2024 · similarity 0.86)
- Towards Characterizing Domain Counterfactuals for Invertible Latent Causal Models (ICLR 2024 · similarity 0.86)
- Counterfactual Contrastive Learning with Normalizing Flows for Robust Treatment Effect Estimation (ICML 2025 · similarity 0.86)
- Bi-Level Decision-Focused Causal Learning for Large-Scale Marketing Optimization: Bridging Observational and Experimental Data (NeurIPS 2025 · similarity 0.85)
- Causal Contrastive Learning for Counterfactual Regression Over Time (NeurIPS 2024 · similarity 0.84)
- A Distribution Mapping Approach to Counterfactually Fair Reinforcement Learning (NeurIPS 2026 · 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.
