Numerical Analysis and Scientific Computing Seminar | Rima Alaifari, Aspects of stability and uniqueness in deep learning algorithms

Zoom (Please contact ddelreyfernandez@uwaterloo.ca for meeting link) 

Speaker

Rima Alaifari, Assistant Professor, Department of Mathematics ETH ETH Zürich

Title

Aspects of stability and uniqueness in deep learning algorithms

Abstract

Deep learning algorithms are nowadays successfully used in different applications including image classification, medical imaging and the solution of PDEs. The first part of this talk will focus on stability aspects of deep neural networks by investigating vulnerability to adversarial attacks. The second part is devoted to operator learning in which we argue that it is favorable to consider a novel concept of neural operators by enforcing a continuous-discrete equivalence, linking operators between infinite-dimensional Hilbert spaces to its discrete realizations