[cs-talks] Upcoming Seminars: IVC (Tues)
fgreen1 at bu.edu
Fri Sep 18 10:44:30 EDT 2015
Geometric optimization for Gaussian Mixture Models
Suvrit Sra, Laboratory for Information and Decision Systems (LIDS)
Tuesday, September 22, 2015 (2-3pm) in MCS 148
Abstract: We take a new look at parameter estimation for Gaussian Mixture Models (GMMs). In particular, we propose using *Riemannian (geometric) manifold optimization* as a powerful counterpart to Expectation Maximization (EM). An out-of-the-box invocation of manifold optimization, however, fails spectacularly: it converges to the same solution but vastly slower. Driven by intuition from manifold convexity, we then propose a reparamerization that has remarkable empirical consequences. It makes manifold optimization not only match EM---a highly encouraging result in itself given the poor record nonlinear programming methods have had against EM so far---but also outperform EM in many practical settings, while displaying much less variability in running times. We further highlight the strengths of manifold optimization by developing a careful manifold LBFGS method that proves even more competitive and reliable than existing manifold optimization tools. We hope that our results encourage a wider consideration of manifold optimization for parameter estimation problems.
The talk will be based on the upcoming NIPS 2015 paper, whose longer version is available at: http://arxiv.org/abs/1506.07677
Suvrit Sra is a Principal Research Scientist (Research Faculty) at the Laboratory for Information and Decision Systems (LIDS) and the Institute for Data, Systems, and Society (IDSS) at MIT since Jan 2015. Prior to that he was a Senior Research Scientist at the Max Planck Institute for Intelligent Systems, in Tübingen, Germany. He obtained Ph.D. in Computer Science from the University of Texas at Austin in 2007. He has previously also held visiting faculty positions at UC Berkeley (EECS) and Carnegie Mellon University (Machine Learning Department). His research is dedicated to bridging a number of mathematical areas such as geometry, analysis, matrix algebra, convex analysis, harmonic analysis, statistics, mathematical optimization, etc., with machine learning, data science, and more broadly the needs of data-driven applications across science and industry.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the cs-talks