Please register your place here.

Speaker:   Professor Jonas Peters, ETH Zurich

Title:           Learning under Change

Abstract:  Many learning methods build on the paradigm that training and test data consist of i.i.d. replicates of the same distribution. In this talk, we discuss when and how this assumption may be violated. We do this from a causal viewpoint and show that such a view suggests how to build stable learning methods. Specifically, we propose stable methods for prediction, dimensionality reduction, and prediction-intervention games. 

The talk does not assume any prior knowledge about causality.

The Florence Nightingale Lecture Series is generously sponsored by XTX Markets.