Deep Learning: Optimization, Generalization and Architectures

Tuesday, 28.5.2019, 14:30
Room 337 Taub Bld.
Tel-Aviv University
Yuval Filmus

Artificial neural networks have recently revolutionized the field of machine learning, demonstrating striking empirical success on tasks such as image understanding, speech recognition and natural language processing. However, we still do not have sufficient theoretical understanding of how such models can be successfully learned. Two specific questions in this context are: how can neural nets be learned despite the non-convexity of the learning problem, and how can they generalize well despite often having more parameters than training data. I will describe some of our results in this context, focusing on the particular properties of common optimization algorithms such as stochastic gradient descent. I will also discuss our recent work on using deep learning for complex-output problems, and describe principles for constructing architectures for these. The resulting models show competitive performance on challenging image understanding tasks such as scene graph generation. Short bio:
Amir Globerson received a BSc in computer science and physics in 1997 from the Hebrew University, and a PhD in computational neuroscience from the Hebrew University in 2006. After his PhD, he was a postdoctoral fellow at the University of Toronto and a Rothschild postdoctoral fellow at MIT. His research interests include machine learning, deep learning, graphical models, optimization, machine vision, and natural language processing. He is an associate editor for the Journal of Machine Learning Research, and was the Associate Editor in Chief for the IEEE Transactions on Pattern Analysis and Machine Intelligence. His work has received several prizes including five paper awards at NeurIPS, ICML and UAI. In 2018 he was the program co-chair for UAI, and general chair for UAI 2019 to be held in Tel Aviv. In 2019 he received the ERC Consolidator Grant.

Back to the index of events