[Ml-stat-talks] [stat seminar reminder]: Friday April 22, 2:30pm by James Xiang

xtong at princeton.edu xtong at princeton.edu
Thu Apr 21 16:18:49 EDT 2011


 PLACE: ORFE 213 (statistics lab) 

TIME: 2:30PM, April 22 

SPEAKER: James Xiang, EE Princeton

TITLE: Beyond Occam's Razor: Classifiers with Structured SparsityABSTRACT:

 Occam's Razor gives preference to the hypothesis that has the simplest
form and explains the data. Indeed, in many machine learning and signal
processing problems, classifiers with sparse parameterized representations
usually show good generalization performances. In this talk, we go beyond
such plain sparsity and look at classifiers with structured sparsity. I
will show that in many cases we are more interested in the structure of
the sparse classifier than merely the sparsity itself. One example is
selecting a sparse yet structured set of voxels (volumetric units in a
brain scan) in fMRI (functional) classification. More importantly, I will
also discuss the theoretical connections between structured sparsity of
decision trees and the sparsity of the wavelet representations of images
and the signal regularity in Besov spaces. Finally I will talk about my
ongoing research in learning large-scale data-adaptive dictionary for
sparse representation. 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.cs.princeton.edu/pipermail/ml-stat-talks/attachments/20110421/cdfbbb70/attachment.htm>


More information about the Ml-stat-talks mailing list