Skip to content Skip to navigation
Seminar
10/1/2014 11:00 am
CoRE A(Room 301)

Adventures in Linear Algebra++ and unsupervised learning

Sanjeev Arora, Princeton University

Organizer(s): Swastik Kopparty and Shubhangi Saraf

Abstract

Many problems in unsupervised learning are NP-complete. To design efficient algorithms with provable guarantees, one must move away from worst-case complexity and make specific assumptions about the input instances. The talk will survey some successes we and others have had in designing such provable algorithms for nononegative matrix factorization, topic modeling, deep learning, hidden markov models, etc. The techniques used in these can be seen as new variants of classical linear algebra: what we think of as Linear Algebra++.