The theory of learning under the uniform distribution is rich and deep. It is connected to cryptography, computational complexity, and analysis of boolean functions to name a few areas. This theory however is very limited in applications due to the fact that the uniform distribution and the corresponding Fourier basis is rarely encountered as a statistical model.

A family of distributions that vastly generalizes the uniform distribution on the Boolean cube is that of distributions represented by Markov Random Fields (MRF). Markov Random Fields are one of the main tools for modeling high dimensional data in all areas of statistics and machine learning. In this paper we initiate the investigation of extending central ideas, methods and algorithms from the theory of learning under the uniform distribution to the setup of learning concepts given examples from MRF distributions. In particular, our results establish a novel connection between properties of MCMC sampling of MRFs and learning under the MRF distribution.

Based on joint work with Elchanan Mossel.


Video Recording