Search papers, labs, and topics across Lattice.
This paper introduces a Deconditional Gaussian Process (DGP) framework for instrumental variable (IV) and proximal causal inference to provide reliable epistemic uncertainty (EU) quantification in the presence of unobserved confounding. The DGP formulation recovers popular kernel estimators as the posterior mean for predictive precision and uses the posterior variance for principled EU quantification. Experiments demonstrate strong predictive performance and informative EU, validated through empirical coverage frequencies and decision-aware accuracy rejection curves.
Gaussian Processes can now provide reliable uncertainty estimates for causal inference with instrumental variables, addressing a critical gap in existing methods.
Instrumental variable (IV) and proximal causal learning (Proxy) methods are central frameworks for causal inference in the presence of unobserved confounding. Despite substantial methodological advances, existing approaches rarely provide reliable epistemic uncertainty (EU) quantification. We address this gap through a Deconditional Gaussian Process (DGP) framework for uncertainty-aware causal learning. Our formulation recovers popular kernel estimators as the posterior mean, ensuring predictive precision, while the posterior variance yields principled and well-calibrated EU. Moreover, the probabilistic structure enables systematic model selection via marginal log-likelihood optimization. Empirical results demonstrate strong predictive performance alongside informative EU quantification, evaluated via empirical coverage frequencies and decision-aware accuracy rejection curves. Together, our approach provides a unified, practical solution for causal inference under unobserved confounding with reliable uncertainty.
