Speaker
Jan M. Pawlowski
Description
In this talk I give a -certainly incomplete- overview of the rapidly growing area of
(i) physics and renormalisation group applications with Machine Learning
as well as
(ii) Machine Learning with the renormalisation group.
Applications (i) use the nonlinear optimisation property of neural networks for physics. Applications
(ii) use the fact that the layerwise or even global information transport in deep neural architectures is
either explicitly or implicitly a general (functional) renormalisation group transformation. Specifically, this
allows us to endow generative architectures which much-needed information beyond the learning sample.
Applications and ideas in (i,ii) are illustrated within simple examples.