12pm-1pm, Wednesday 27th September 2023
Presented by Prof Geordie Williamson
Venue: Carslaw Seminar Room 351, F07
Abstract: In many applications of AI, one only cares about getting the answer right (or right most of the time). If a self-driving car doesn’t crash we are happy. However, for applications in mathematics (and science more broadly) one cares much more about the “why?”. Interpretability is a massive issue in modern AI. Since 2020 I have been working with DeepMind on applications of machine learning in mathematics. I’ll try to discuss some success stories, and the (major) hurdles that remain.