Reproduce “Density Ratio-based Proxy Causal Learning Without Density Ratios” 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
Bariscan Bozkurt, Ben Deaner, Dimitri Meunier, Liyuan Xu, Arthur Gretton
The paper presents proxy causal learning for estimating treatment effects when a confounder is hidden and treatment and outcome proxies are observed. It constructs a treatment bridge with kernel ridge regression, avoiding explicit density-ratio estimation. On dSprite’s 4,096-dimensional treatments, error is lower than Kernel Negative Control and comparable to KPV and PMMR, under stated identification assumptions.
Key insights
- The least-squares objective for the treatment bridge function can be algebraically simplified to eliminate the need for explicit density ratio estimation.
- Utilizing Reproducing Kernel Hilbert Spaces enables the derivation of closed-form solutions for causal effect estimators through multi-stage kernel ridge regression, enhancing computational tractability.
- Ablation studies reveal that treatment bridge methods can perform better when the outcome proxy is more informative, while outcome bridge methods benefit from a more informative treatment proxy, suggesting complementary strengths.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.