2022-03-21 · 16:00 UTC
Connections between deep learning and Gaussian process methods are emerging as a tool for understanding deep learning. The kernels involved are “spherically stationary”: in this talk I’ll illustrate where these spheres come from, and talk about how the spherical structure can be used for efficient inference in the Gaussian process analogues of wide deep networks.
James Hensman is a Principle Applied Scientist at Amazon, where he works on machine learning models as part of a digital twin that is used to control and plan Amazon’s supply chain. Previously James was Director of Research at Secondmind.ai, and prior to this he was a University Lecturer at Lancaster University. His research interests centre on Bayesian machine learning methods and Gaussian processes, and he has applied his knowledge in fields from Biostatistics to Big tech.