Causal inference with fewer assumptions

Kirtan Padh (Ph.D. Student)

Causality has recently been seen as a promising path towards designing more reliable machine learning algorithms, in the sense of robustness, domain generalization, fairness, interpretability and more. However, there have also been some criticisms regarding the strong and unrealistic assumptions that causal inference methods often rely on. It is natural that stronger assumptions allow for stronger inference. However, the trade-off between the strength of assumptions and strength of inference is arguably a continuum. The goal of this thesis is to provide methods which allow practitioners to choose a point on this continuum they feel most comfortable with in terms of the assumptions they are willing to make, and the strength of the inference will be adjusted accordingly. Hopefully this helps us take a step towards fulfilling the promise of causality and its application in the real world.

Primary Host:	Niki Kilbertus (Technical University of Munich & Helmholtz Center Munich)
Exchange Host:	Ricardo Silva (University College London)
PhD Duration:	01 March 2021 - 28 February 2025
Exchange Duration:	- Ongoing - Ongoing