Supervisors

Suitable for

Abstract

This project deals with the fundamental problem of counterfactual estimation. Counterfactual estimation (CE) refers to the process of estimating or predicting what would have happened in a given situation if different actions or events had taken place. Counterfactual estimation is particularly vital in healthcare due to the complex and interconnected nature of biological systems and the need to understand the true causes behind diseases, treatment effects, and patient outcomes.

Although CE holds tremendous promise and potential, estimating counterfactuals remains a significant challenge. Recently, transformer-type algorithms have emerged as a powerful tool for the related problem of zero-shot treatment effect estimation. In this project, we will look into transformer-type algorithms for CE. The work will consist of: (i) reading and understanding [1]; (ii) extend the framework of [1] to counterfactual estimation from interventional data; (iii) implement and perform experiments with a base model for this extended framework.

[1] Towards Causal Foundation Model: on Duality between Causal Inference and Attention. Jiaqi Zhang et al. CoRR abs/2310.00809 (2023)

Pre-requisites: A student suitable for this project will work with causal inference, attention mechanisms, and possibly the basics of reproducing kernel Hilbert spaces and feature maps. A background on any of these topics is desirable. Coding skills are required.