Speaker: Jonas Peters, KU

Title: One goal, two fields: Learning causality from invariance in machine learning and in econometrics

Abstract: We study the problem of learning a causal relationship between predictors X and a response Y from data. To do so, we exploit the existence of an exogenous variable and consider models that are invariant, rather than prediction optimal. Similar aspects of this question have been studied both in the econometrics literature and in the machine learning and statistics literature. In this talk, we present a few recent results on invariance, causality and distributional robustness and aim to relate them to classical results in econometrics. We believe that bridging the gap between causality research in econometrics and machine learning and statistics would be beneficial for both fields.

Area of research: causality, computational statistics, machine learning, robustness, independence testing

Host: Allan Würtz

Organisers: Luke Taylor and Mikkel Bennedsen