Computational Neurobiology Laboratory at the Salk Institute

NIPS 2004 Workshop

December 17, 2004
Whistler, B.C., Canada

Overcomplete Representations

 

Organizers:
Terry Sejnowski,The Salk Institute and University of California at San Diego, USA
Steve Zucker,Yale University, Connecticut, USA

http://www.ahebbifs.org/nips2004/

The problem of finding sparse representations in overcomplete bases
has many applications in signal processing and is an area of intense
research. Overcomplete representations can be more sparse than
complete respresentations and have greater flexibility in matching
the structure in the data. In some circumstances the overcomplete
codes can have great robustness to noise. Overcomplete codes have
also been proposed as a model of some of the response properties of
neurons in primary visual cortex, which is x100 overcomplete compared
to the number of input fibers from the retina.

This workshop will explore a range of topics including:

- Algorithms for computing overcomplete representations (matching
pursuit, ICA, etc.)
- Learning (e.g., ICA) vs designing (e.g., curvelets) overcomplete
representation systems.
- Analysis and performance of techniques.
- Defining optimality criteria (e.g. L^0 norm or L^1 norm sparsity?)
- Natural statistics (images, sound, etc.) and their implications for
overcomplete representations.
- Applications, e.g., coding, classification, or inference using
overcomplete representations.

Friday morning session: 7:30am-10:30am

7:30
Methods for dictionary learning and component estimation
Ken Kreutz-Delgado, UCSD

8:00am
(Matching, Orthogonal, Basis) pursuit and the foundations of linear programming
Dan Spielman, MIT

8:30am
A geometric view of overcomplete representations
Patrick Huggins, Yale University

9:00am coffee break

9:30
TBA
Dana Ballard, Rochester University

10:00am
Robust coding over noisy channels with sparse overcomplete representations
Eizaburo Doi, CMU

Friday afternoon session: 4:00pm-7:00pm

4:00pm
Energy-based models for overcomplete representations
Simon Osindero, Univ of Toronto

4:30pm
Learning shiftable basis functions from natural images
Bruno Olshausen, Redwood Institute and UC Davis

5:00
Overcompleteness, the partition function, time and spikes
Tony Bell, Redwood Institute

5:30pm coffee break

6:00
Explicit invariance as an alternative to pursuit in overcomplete representations
Eero Simoncelli, NYU

6:30
Assignment of multiplicative mixtures in natural Images
Odelia Schwartz, Salk Institute