Geordie Williamson

Abstract: Finding proofs and finding examples are critical parts of the mathematical process. Over the last few years there has been enormous effort in academia and industry to try to use AI to discover proofs. The focus of my talk will be different: can we use AI to find interesting examples?

I will focus on three different areas where AI has proven useful: extremal combinatorics, combinatorial geometry and knot theory. I will explain the role played by the neural network in each search task. A common theme is that although the neural network improves search in each example (and in several cases allows one to discover the best-known examples), the effect is often subtle and interesting. By looking at several examples in parallel, I hope to give a sense of where neural networks might help the mathematician interested in finding new phenomena using computers and AI.

This is joint work with lots of people: the extremal combinatorics work is joint with Charton, Ellenberg and Wagner. The knot theory work is joint with Charton, Narayanan and Yacobi, and the combinatorial geometry work is joint with Davies, Gupta, Racanière, Swirszcz, Weber and Wagner.

Click here to convert to your local time: 8th May 9:00 CET