Causality in
Medical Image Computing

ABOUT THE SPEAKERS

Fabio De Sousa Ribeiro

Abstract: The ability to generate plausible counterfactuals has wide scientific applicability and is particularly valuable in fields like medical imaging, wherein data are scarce and underrepresentation of subgroups is prevalent. Answering counterfactual queries such as 'why?' and 'what if?' in the language of causality could greatly benefit key research areas, including: (i) explainability; (ii) data augmentation; (iii) robustness to spurious correlations; and (iv) fairness in both observed and hypothetical outcomes.

Despite recent advancements, accurate estimation of interventional and counterfactual queries for high-dimensional structured variables like medical images remains an open problem. Evaluating inferred counterfactuals also poses inherent challenges, as they are unobservable by definition. In this talk, we will present a system and method for developing and evaluating deep causal mechanisms for structured variables. Leveraging insights from causal mediation analysis and advances in deep generative modeling, we will show how such deep causal mechanisms are capable of plausible counterfactual inference, as measured by the axiomatic soundness of counterfactuals.

Bio: Fabio is a postdoctoral research associate in the BioMedIA group at Imperial College London. His primary research interests lie at the intersection of causality and deep generative modelling for medical imaging and healthcare applications. His work bolsters the ongoing effort by the machine learning community to combine the central ideas behind causality and deep representation learning to help tackle several challenging research areas such as explainability, robustness and fairness.

Maggie Makar

Abstract: Robustness to certain forms of distribution shift is a key concern in many ML applications. Often, robustness can be formulated as enforcing invariances to particular interventions on the data generating process. In this talk, I will discuss causally-motivated approaches to enforcing such invariances, paying special attention to shortcut learning, where a robust predictor can achieve optimal i.i.d generalization in principle, but instead it relies on spurious correlations or shortcuts in practice. I will discuss approaches which utilize auxiliary labels, typically available at training time, to enforce conditional independences between the latent factors that determine these labels. I will focus on settings where the shortcuts are known a priori and when the shortcuts are unknown, and highlight important theoretical properties of these causally motivated training techniques commonly used in causal inference, fairness, and disentanglement literature.

Bio: Maggie is an assistant professor of computer science and engineering at the University of Michigan in Ann Arbor. She completed her PhD in computer science at MIT, and has held research positions at Google Brain and Microsoft. Dr. Makar's work is funded by the NSF and the Schmidt Futures foundation.