ICONIP-2000 Special Session:
Models of Natural Image Statistics

Organizers: Aapo Hyvärinen, Mark Girolami, Patrik Hoyer, and Te-Won Lee
 

Recently, a lot of interest has been given to models attempting to learn representations from natural images using statistical models. This special session groups together some of the most important researchers on this topic and closely related topics such as image feature extraction and V1 function. Contributions cover the range from low-level processing to high-level pattern recognition.
 

List of invited speakers and talk titles (in no particular order):

Daniel Lee
Bell Labs, Lucent Technologies, USA
"Algorithms and applications of non-negative matrix factorization"

Mike Lewicki
CNBC, Carnegie Mellon University, USA
"Efficient coding of natural images"

Norbert Mayer
Max Planck Institut für Strömungsforschung, Germany
"The impact of receptive field shape on cortical map formation"

Jong-Hoon Oh
Postech, Korea
"Learning sparse code for natural images using on-line up-propagation algorithm"

Aapo Hyvärinen
Helsinki Univ. of Tech., Finland
"Topographic ICA as a model of V1 receptive fields and topography"

Patrik Hoyer
Helsinki Univ. of Tech., Finland
"Modeling chromatic and binocular properties of V1 topography using topographic ICA"

Te-Won Lee
Institute for Neural Computation at UCSD, USA
"The chromatic structure of natural scenes"

Erkki Oja
Helsinki Univ. of Tech., Finland
"Analyzing low-level visual features using content-based image retrieval"

Shigeru Tanaka
RIKEN, Japan
"Nonlinear binocular integration for motion-in-depth representation in the cat primary visual cortex"