Reliable ML: from LLMs to cyber-physical systems

Interacting systems - €”ranging from biological networks and social webs to critical infrastructures like power grids - €”pose distinctive modeling challenges. Unlike many physical systems with well-established governing equations, most interacting systems lack explicit dynamical laws, making data-driven modeling or machine learning essential. Yet, standard machine learning methods often break down under distribution shifts or high noise, and struggle to provide reliable predictions in the face of unexpected or rare scenarios. The aim of this summer school is to address core questions related to reliable learning under noise, distributional shifts, uncertainty and beyond. The lectures will be given by an outstanding panel of speakers from academia (ETHZ, EPFL, MIT, Northwestern) and industry (Apple, DeepMind, Isomorphic Labs, ...).