Reproduce “Cross-Head Attention Uplift Network with Inverse Propensity Score under Unobserved Confounding” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Haoran Zhang, Chuanpu Li, Yuxin Fu, Bin Tong, Guan Wang, Bo Zheng, Feng Zhou
CHAUN uses attention-based interactions between treatment and control representations to estimate uplift, while RA-IPS tests outcome learning under bounded changes to nominal treatment-assignment probabilities. On LAZADA, its normalized QINI is 0.0314, versus DragonNet’s 0.0250; on LAZADA-Masked, RA-IPS raises QINI from 0.0221 to 0.0230. These five-seed results support better ranking in evaluated settings, not recovery of counterfactuals or elimination of hidden confounding.
Key insights
- Individual Treatment Effects remain identifiable even in the presence of unobserved confounders, provided the true propensity scores are known and used for reweighting.
- Dynamic cross-head attention in the outcome learner effectively balances the exploitation of inter-group similarities with the differentiation of treatment-specific effects, enhancing ITE estimation, especially in complex, high-dimensional data.
- Adversarial propensity score reweighting under unobserved confounding requires a refined uncertainty set with global statistical constraints to avoid trivial solutions and ensure robust debiasing.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.