ColloquiaSanjoy Dasgupta - talk on Monday, February 25th

[Margery Ishmael]

Department of Computer Science/University of Chicago

1100 E. 58th Street, Ryerson Hall

Talk Announcement

Monday, February 25th at 2:30 p.m. in Ryerson 251

SANJOY DASGUPTA, AT&T LABS-RESEARCH

Title: "Learning mixtures of Gaussians"

Abstract: One of the most common ways to cluster data is to model it as a 
mixture of Gaussians.  This is the most well-weathered statistical model of 
clustered data, and has been the subject of close study for well over a 
century.  However, the only algorithms known for this clustering task have 
been heuristics with rather weak performance guarantees.

I shall describe the first efficient, provably correct algorithm to learn a 
mixture of Gaussians from arbitrarily high-dimensional data.  It recovers 
the centers of the Gaussian clusters to within the precision specified by 
the user, in time which scales linearly in the dimension of the data and 
polynomially in the number of Gaussians.

The algorithm depends crucially upon an understanding of the geometric 
structure of high dimensional Gaussian clusters, and upon a particular 
projection technique which reduced the dimension of the data while at the 
same time probably enhancing useful aspects of this structure.

http://www.research.att.com/~dasgupta/

*The talk will be followed by refreshments in Ryerson 255*
Persons with disabilities who may need assistance, please call 773.834.8977



=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Margery Ishmael
Secretary to the Chairman, Department of Computer Science
The University of Chicago
tel. 773.834.8977  fax. 773.702.8487