Machine Learning at the Edge of Ignorance

The lecture describes how machine learning has transformed classical, precisely defined computation into data-driven learning and generalization to new situations. Today, it is a key technology for scientific progress, creative applications, and societal transformation. At the same time, fundamental weaknesses such as a lack of robustness, hallucinations, or problems with unknown data become apparent in complex operating environments. These challenges highlight the difficulty of reliably working at the boundary between known and unknown knowledge. This presentation introduces approaches that enable AI systems to recognize, analyze, and make transparent their knowledge boundaries in order to facilitate better decision-making.

Krikamol Muandet is a senior scientist at CISPA and heads the Rational Intelligence Lab there, having previously worked at the Max Planck Institute for Intelligent Systems and Mahidol University, among other institutions. His research focuses on statistical learning theory, kernel methods, causal inference, counterfactual prediction, and algorithmic game theory.