Estimating Causal Effects using a Multi-task Deep Ensemble

TL;DR

This paper introduces CMDE, a deep ensemble framework for estimating causal effects in complex data like images, demonstrating superior performance and uncertainty quantification over existing methods.

Contribution

The paper presents CMDE, a novel multi-task deep ensemble approach that effectively handles high-dimensional, multi-modal data for causal effect estimation, with theoretical and empirical validation.

Findings

CMDE outperforms state-of-the-art methods on various datasets.

It provides reliable pointwise uncertainty estimates.

Theoretical proofs establish equivalence to multi-task Gaussian processes.

Abstract

A number of methods have been proposed for causal effect estimation, yet few have demonstrated efficacy in handling data with complex structures, such as images. To fill this gap, we propose Causal Multi-task Deep Ensemble (CMDE), a novel framework that learns both shared and group-specific information from the study population. We provide proofs demonstrating equivalency of CDME to a multi-task Gaussian process (GP) with a coregionalization kernel a priori. Compared to multi-task GP, CMDE efficiently handles high-dimensional and multi-modal covariates and provides pointwise uncertainty estimates of causal effects. We evaluate our method across various types of datasets and tasks and find that CMDE outperforms state-of-the-art methods on a majority of these tasks.