Francis Bach

Francis Bach is a Senior Research Scientist at Inria in Paris, where he leads the SIERRA project-team, which is part of the computer science department of Ecole Normale Supérieure. His research spans machine learning theory, optimization, kernel methods, neural networks, uncertainty calibration, decision-making, and algorithms with guarantees. He is a member of the French Academy of Sciences and has received major distinctions in machine learning and optimization, including test-of-time and best-paper awards at leading conferences. He is the author of the book “Learning Theory from First Principles” and has supervised a large international group of students and postdoctoral researchers. His current work focuses on optimization, reliable learning, and decision-making under uncertainty.

AI2050 Project

AI systems increasingly act: they call tools, plan experiments, control robots, and make recommendations. Yet we lack a mature science for deciding when an AI action is reliable. Bach’s project will build an external technical auditing layer. From observed behavior and controlled tests, it will learn compact descriptions of situations and actions, plus sets of objectives compatible with the evidence. The auditor will combine them with externally declared safety constraints to report when an action is supported, when more information is needed, and when the system should defer.