Approximate Manifold Sampling

Abstract

Sampling from a probability density concentrated around a lower-dimensional submanifold is of importance in numerous applications arising in machine learning, statistics and statistical physics. This task is particularly challenging due to the extreme anisotropy and high-dimensionality of the problem, and the correlation between the variables. We propose a new family of bespoke algorithms to sample efficiently from these densities and show their computational superiority to general purpose and specialized samplers. In particular, we show that in some cases it is able to efficiently sample from densities that are several orders of magnitude tighter around the submanifold.

Mauro Camara Escudero

Research Associate in Statistical Machine Learning

My research interests include approximate manifold sampling and generative models.