Feedforward Neural Networks for Quantum Frustrated Magnets

Representation of an arbitrary quantum mechanical wave-function requires an exponential amount of memory. A key question in the field, then, is to understand whether there are interesting classes of wave-functions which can be compactly represented. In this talk, we explore this question in the context of feedforward neural networks. We numerically study the expressivity of the feedforward artificial neural networks as variational ansatz for ground states of frustrated magnetic systems, investigating the scaling of the number of variational parameters with the system size as well as the effect of different network architectures.