BEGIN:VCALENDAR
CALSCALE:GREGORIAN
METHOD:PUBLISH
VERSION:2.0
PRODID:-//PlanIt Purple//EN
BEGIN:VEVENT
STATUS:CONFIRMED
LAST-MODIFIED:19691231T180000
URL:
PRIORITY:0
CLASS:PUBLIC
UID:431049@northwestern.edu
SUMMARY:IEMS Seminar - Hogwild for Machine Learning on Multicore
DESCRIPTION:Hogwild for Machine Learning on Multicore    Ben Recht       University of Wisconsin-Madison    Abstract: Stochastic Gradient Descent (SGD) is a popular optimization algorithm for solving data-driven machine learning problems such as classification\, model selection\, sequence labeling\, and recommendation.  SGD is well suited to processing large amounts of data due to its  robustness against noise\, rapid convergence rates\, and predictable memory footprint. Nevertheless\, SGD seems to be impeded by many classical barriers to scalability: (1) SGD appears to be inherently sequential\, (2) SGD assumes uniform sampling from the underlying data set resulting in poor locality\, and (3) current approaches to parallelize SGD require performance-destroying\, fine-grained communication. This talk aims to refute the conventional wisdom that SGD inherently suffers from these impediments. Specifically\, I will show that SGD can be implemented in parallel with minimal communication\, with no locking or synchronization\, and with strong spatial locality.  I will provide both theoretical and experimental evidence demonstrating the achievement of linear speedups on multicore workstations on several benchmark optimization problems. Finally\, I will close with a discussion of a challenging problem raised by our implementations relating arithmetic and geometric means of positive definite matrices. Joint work with Feng Niu\, Christopher Re\, and Stephen Wright.   Biography: Benjamin Recht is an Assistant Professor in the Department of Computer Sciences at the University of Wisconsin-Madison and holds courtesy appointments in Electrical and Computer Engineering\, Mathematics\, and Statistics.  He is a PI in the Wisconsin Institute for Discovery (WID)\, a newly founded center for research at the convergence of information technology\, biotechnology\, and nanotechnology. Ben received his B.S. in Mathematics from the University of Chicago\, and received a M.S. and PhD from the MIT Media Laboratory.  He is the recipient of an NSF Career Award and an Alfred P. Sloan Research Fellowship 
DTSTART:20120508T160000
DTEND:20120508T170000
CREATED:20120502T000000
DTSTAMP:20120502T000000
SEQUENCE:0
LOCATION:Evanston
END:VEVENT
END:VCALENDAR