When: Thursday 6th of April, 1pm AEST
Where: This seminar will be partially presented at the Rose Street Seminar area (J04) and partially online via Zoom
Speaker: Dr Ruigang Wang
Title: Direct Parameterization of Lipschitz-Bounded Deep Networks
In this talk I will present a new parameterization of deep neural networks (both fully connected and convolutional) with guaranteed Lipschitz bounds, i.e. limited sensitivity to perturbations. The Lipschitz guarantees are equivalent to the tightest-known bounds based on certification via a semidefinite program (SDP), which does not scale to large models. In contrast to the SDP approach, we provide a “direct” parameterization, i.e. a smooth mapping from a Euclidean space onto the set of weights of Lipschitz-bounded networks. This enables training via standard gradient methods, without any computationally intensive projections or barrier terms. The new parameterization can equivalently be thought of as either a new layer type (the sandwich layer), or a novel parameterization of standard feedforward networks with parameter sharing between neighbouring layers. We illustrate the method with some applications in image classification.
Ruigang Wang received his Ph.D. degree from UNSW in 2017. From 2017 to 2018 he worked as a postdoc at UNSW, focusing on distributed model predictive control for large-scale systems. Then, he works as a postdoc at ACFR since 2018. His research interests include robust machine learning and nonlinear control theory.