BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Maximum Entropy and Applications in Natural and Social Sciences
DTSTART:20090427T161500
DTSTAMP:20260511T084301Z
UID:7b5959d5de54bf9c76c496e11b2291f83e89bc979dbffe5503209b3d
CATEGORIES:Conferences - Seminars
DESCRIPTION:Dr. Miroslav Dudik\, Carnegie Mellon University\, USA\nThe max
 imum entropy approach (maxent)\, equivalent to maximum likelihood\, is a w
 idely used density-estimation technique. However\, when trained on small d
 atasets\, maxent is likely to overfit\, and when trained over large sample
  spaces\, naive implementations of maxent are intractable. To prevent over
 fitting\, we propose a relaxed version of maxent\, which turns out to be e
 quivalent to L1-regularized log likelihood. We prove strong statistical gu
 arantees for L1-regularized maxent\, and show how it can be generalized to
  the problem of estimation in the presence of sample-selection bias\, and 
 to the problem of simultaneous estimation of multiple densities. To addres
 s the computational challenges\, we propose an approach based on sampling 
 and coordinate descent.\n\nI discuss two applications of maxent: modeling 
 distributions of biological species and modeling cross-cultural negotiatio
 n\, focusing mainly on the former. Regularized maxent fits species distrib
 ution modeling well and offers several advantages over previous techniques
 .\nIn particular\, it addresses the problem in a statistically sound manne
 r and allows principled extensions to situations when the data-collection 
 process is biased or when we have access to data on many related species. 
 Throughout the talk I will demonstrate the benefits of our approach on lar
 ge\, real-world modeling problems.\n\nBased on joint work with Rob Schapir
 e\, Steven Phillips\, Geoff Gordon\, Dave Blei and others.\n\nBio: Mirosla
 v Dudik received his PhD in Computer Science from Princeton University in 
 2007. Currently\, he is a postdoctoral researcher at Carnegie Mellon Unive
 rsity. His interests are in theoretical and applied aspects of machine lea
 rning\, both statistical and algorithmic. He focuses on small-sample densi
 ty estimation and game-theoretic modeling.\n\nM. Dudik's homepage
LOCATION:INM202
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR