From PDE-Constrained Optimization to DNNs

dc.contributor	Mang, Andreas
dc.contributor.author	Ali Hamza Abidi, Syed
dc.date.accessioned	2022-09-22T20:55:41Z
dc.date.available	2022-09-22T20:55:41Z
dc.date.issued	2022-04-14
dc.description.abstract	In the present work we explore numerical methods inspired by optimal control theory to train image classifiers. In a first step, we consider a prototypical formulation of a variational optimization problem governed by an elliptic dynamical system. We will discuss the numerical treatment and study some of the mathematical operators. Subsequently, we present the optimal control formulation for training DNNs and derive some expressions for the associated optimality conditions. In our future work, we plan to extend these optimality conditions and device a numerical scheme for the DNN training problem, similar to the scheme developed for the prototypical problem.
dc.description.department	Mathematics, Department of
dc.description.department	Honors College
dc.identifier.uri	https://hdl.handle.net/10657/11716
dc.language.iso	en_US
dc.relation.ispartof	Summer Undergraduate Research Fellowship
dc.rights	The author of this work is the copyright owner. UH Libraries and the Texas Digital Library have their permission to store and provide access to this work. Further transmission, reproduction, or presentation of this work is prohibited except with permission of the author(s).
dc.title	From PDE-Constrained Optimization to DNNs
dc.type	Poster

Files

Now showing 1 - 1 of 1