Speaker
Description
Applications of machine learning techniques to numerical studies of quantum field theories have been explored intensely in recent years. One such application is the use of a neural network for finding a map between the Boltzmann distribution of a lattice field theory and a simpler distribution function (a 'trivializing map' or 'normalizing flow'). Once such a map is found, one expects to improve the Monte-Carlo simulation by sampling field configurations from the simpler distribution function. In this talk, I will discuss the application of normalizing flows to $\phi^4$ real scalar field theory in Minkowski space as a first step toward solving its real-time sign problem. The goal is to find a map between the complex-valued Boltzmann distribution via the action of $\phi^4$ theory and a real-valued distribution. Firstly I will explain the conjectured existence of such normalizing flows which solve the sign problem completely. Then I will discuss the search for such normalizing flows with the aid of machine learning and the perturbative study of normalizing flows.
