Statistical Mechanics Approach to Inverse Problems on Networks

Abstract

Statistical Mechanics has gained a central role in modern Inference and Computer Science. Many optimization and inference problems can be cast in a Statistical Mechanics framework, and various concepts and methods developed in this area of Physics can be very helpful not only in the theoretical analysis, but also constitute valuable tools for solving single instance cases of hard inference and computational tasks. In this work, I address various inverse problems on networks, from models of epidemic spreading to learning in neural networks, and apply a variety of methods which have been developed in the context of Disordered Systems, namely Replica and Cavity methods from the theoretical side, and their algorithmic incarnation, Belief Propagation, to solve hard inverse problems which can be formulated in a Bayesian framework