UTCS FACULTY CANDIDATE: Allen Yang/University of California-Berkeley: "High-Dimensional Multi-Model Estimation -- Its Algebra, Statistics, and Sparse Representation" ACES 2.302 Thursday, April 9, 2009 11:00 a.m.

Contact Name: 
Jenna Whitney
Date: 
Apr 9, 2009 11:00am - 12:00pm

There is a sign up schedule for this event:

http://www

.cs.utexas.edu/department/webevent/utcs/events/cgi/eidshow.cgi?person=Allen

Yang-FACULTYCANDIDATE

Type of Talk:  UTCS FACULTY CANDIDATE

Speaker/Affiliation:  Allen Yang/University of California-Berkeley

Date/Time:  Thursday, April 9, 2009  11:00 a.m.

Location:  ACES 2.302

Host:  Kristen Grauman

Talk T

itle: "High-Dimensional Multi-Model Estimation -- Its Algebra, Stati

stics, and Sparse Representation"

Talk Abstract:

 

Recent advances in information technologies have led to unprecedented large
amounts of high-dimensional data from many emerging applications. The need
for more advanced techniques to analyze such complex data calls for shifti

ng research paradigms. In this talk, I will overview and highlight several
results in the area of mixture-model estimation in high-dimensional data s

paces. Applications will be presented such as motion segmentation, image s

egmentation, face recognition, and human action categorization. Through t

his talk, I intend to emphasize the confluence of algebra and statistics t

hat may lead to more advanced solutions in analyzing complex, singular dat

a structures such as mixture linear subspaces and nonlinear manifolds.

In the first part of the talk, I will introduce an algebro-geometri

c technique to simultaneously segment mixture linear and nonlinear manifold

s, and applications in image-based motion segmentation and texture segment

ation. The solutions are robust the moderate data noise and outliers, and

outperform classical model-estimation methods such as RANSAC and Normalized

-Cut. The second part will be focused on classification of mixture subspace
models and its application in face recognition, where the prior informati

on about the subspaces is provided through training examples. Inspired by c

ompressive sensing theory, the recognition problem is reformulated via a s

parse representation. Furthermore, efficient solutions exist to recover su

ch sparse representation under high data distortion using L-1 minimization.
Finally, I will discuss several open problems in the emerging field of di

stributed sensor perception.