Title: Adventures in Linear Algebra++ and unsupervised learning
Speaker: Sanjeev Arora, Princeton University
Date: Wednesday, October 1, 2014 11:00-12:00pm
Location: CoRE Bldg, Room 301A, Rutgers University, Busch Campus, Piscataway, NJ
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++.
See: http://www.math.rutgers.edu/~sk1233/theory-seminar/S14/