A Picture of the Energy Landscape of Deep Neural Networks

Speaker: Pratik Chaudhari
Affiliation: UCLA - Computer Science Dept.

Abstract: Deep networks are mysterious. These over-parametrized machine learning models, trained on a handful of data, with rudimentary optimization algorithms, on non-convex landscapes, in millions of dimensions, have hitherto defied attempts to put a sound theoretical footing beneath their impressive performance.

This talk will shed light upon some of these mysteries. I will employ diverse ideas — from thermodynamics and optimal transportation, to partial differential equations, control theory and Bayesian inference — and paint a picture of the training process of deep networks. Along the way, I will develop state-of-the-art algorithms for non-convex optimization.

In a broader perspective, an autonomous system entails, not just the ability to classify objects in images but also, active perception: the ability to guide the data acquisition process, and control: the ability to act upon this data and seamlessly manipulate the physical world. I will conclude with a vision of how advances in machine learning and robotics may come together to help build such a Cyber-Physical Intelligence.

Biography:  Pratik Chaudhari is a PhD candidate in Computer Science at UCLA where he works with Stefano Soatto. His research interests include deep learning, robotics and computer vision. He has worked on perception and control algorithms for autonomous urban navigation as a part of nuTonomy Inc. Pratik holds Master’s and Engineer’s degrees from MIT and a Bachelor’s degree from IIT Bombay in Aeronautics and Astronautics.

For more information, contact Prof. Paulo Tabuada (tabuada@ucla.edu)

Date(s) - Feb 05, 2018
11:00 am - 12:30 pm

EE-IV Shannon Room #54-134
420 Westwood Plaza - 5th Flr., Los Angeles CA 90095