Speaker
Description
Neural Quantum States started as a machine-learning method to solve for ground states of quantum systems [1], and is now a fully-fledged framework which can deliver competent results when compared to the standard many-body methods. At its core, it is Variational Monte Carlo, with the only difference that the ansätze are neural networks. Therefore, the central calculation is energy minimization, and numerical optimizers are core algorithms within this method. While the most widely used optimizer is Stochastic Reconfiguration [2], it has been recently shown that other optimization techniques, namely Decisional Gradient Descent (DGD) [3], can beat it in some scenarios. In this talk, I will cover the basic idea behind DGD, and explain how we are extending its reach beyond continuous degrees of freedom to encompass spin systems as well, with an outlook on atomic nuclei.
[1] G. Carleo and M. Troyer (2017)
[2] F. Becca and S. Sorella (2017)
[3] M. Drissi, J. Keeble, J. Rozalén Sarmiento, A. Rios (2024)