- Leonardo Zepeda-Nunez, The University of California, Berkeley
Deep learning has rapidly become a large field with an ever-growing range of applications; however, its intersection with scientific computing remains in its infancy, mainly due to the high accuracy that scientific computing problems require, which depends greatly on the architecture of the neural network. The speaker will present a novel deep neural network with a multiscale architecture inspired in H-matrices (and H2-matrices) to efficiently approximate, within three or four digits, several challenging nonlinear maps arising from the discretization of PDEs, whose evaluation would otherwise require computationally intensive iterative methods. The talk will focus on the notoriously difficult Kohn-Sham map arising from density functional theory. The speaker will show (1) that the proposed multiscale-neural network can efficiently learn this map, thus bypassing an expensive self-consistent field iteration and (2) the application of this methodology to ab initio molecular dynamics, for which examples for 1D problems and small, albeit realistic, 3D systems will be provided.
About the Speaker:
Leonardo Zepeda-Núnez is an Postdoctoral Researcher at UC Berkeley and Lawrence Berkeley National Laboratory.