Maithra Raghu

My research at Brain has so far focused on scientifically understanding aspects of deep neural networks. This has included

• Exploring Neural Network Expressivity: how the type of neural network architecture used affects the kinds of functions it can compute. Papers in Exponential Expressivity in Deep Neural Networks through Transient Chaos (at NIPS 2016) and On the Expressive Power of Deep Neural Networks (ICML 2017).
• Probing the representations learned by deep models and understanding learning dynamics SVCCA (at NIPS 2017).

Other work I have done has looked at the algorithmic underpinnings of group selection, Team Peformance with Test Scores (at EC 2015) and a lightweight sequence prediction model Linear Additive Markov Processes (at WWW 2017).