Harold Erbin: Machine learning for Riemann surfaces and applications to string field theory


String field theory is a second-quantized formulation of string theory. While its general properties are well understood, its action is non-polynomial and requires the determination of complicated subspaces of the moduli spaces of Riemann surfaces (vertex regions) and some specific conformal maps. An elegant parametrization is provided by minimal area metrics built from Strebel differentials on n-punctured spheres, but it requires solving the notoriously difficult accessory parameter problem. In this talk, I will describe recent works where we construct the necessary data for the quartic interaction in terms of neural networks. As a consistency check, we recover the known quartic term in the closed string tachyon potential. I will also argue that the method generalizes to higher orders.