Seminar, Hosted by London Mathematical Society
LMS Computer Science Colloquium 2026: Theory and Practice of Machine Learning
Francois-Xavier Briol, Mauricio Álvarez, Özgür Şimşek and Adrián Javaloy
University College London; University of Manchester; University of Bath
Abstract
Four talks connect learning theory to practical inference. The programme covers inexpensive global sensitivity analysis for Bayesian priors, scalable Gaussian-process variational autoencoders using nearest neighbours and inducing variables, and graph-based construction of temporally extended actions in reinforcement learning. A fourth talk studies generative models for interventions and counterfactuals, including hidden confounding and the use of proxy variables.