Visual cortex dissertation (with software and web demos)

[James A. Bednar]

I am pleased to announce the availability of my Ph.D. dissertation,
completed last year at the Department of Computer Sciences at the
University of Texas at Austin under the supervision of Prof. Risto
Miikkulainen.  The thesis shows how the combination of spontaneous
neural activity and visual patterns in the environment can help
explain the development of orientation and face processing circuitry
in the cortex.  Early versions of this work appeared at the AAAI-2000,
CogSci-2002, and CNS*02 conferences, and updated results will appear
in Neural Computation, 2003 (in press).

Jefferson Provost and I are also pleased to announce version 4.0 of
the LISSOM software package for self-organization of laterally
connected maps in visual cortex.  This software supports the
simulations in the dissertation, and is intended to serve as a
starting point for computational studies of the development and
function of perceptual maps in general.

The software is available at topographica.org, and the dissertation
and other papers and demos are available at
http://www.cs.utexas.edu/users/nn/pages/research/visualcortex.html.

-- James A. Bednar


_______________________________________________________________________________
Publications:

LEARNING TO SEE: GENETIC AND ENVIRONMENTAL INFLUENCES ON VISUAL DEVELOPMENT
James A. Bednar 
Ph.D. Thesis, The University of Texas at Austin
Technical Report~AI-TR-02-294, May 2002 (138 pages).

http://www.cs.utexas.edu/users/nn/pages/publications/abstracts.html#bednar.phd02

How can a computing system as complex as the human visual system be
specified and constructed? Recent discoveries of widespread
spontaneous neural activity suggest a simple yet powerful explanation:
genetic information may be expressed as internally generated training
patterns for a general-purpose learning system. The thesis presents an
implementation of this idea as a detailed, large-scale computational
model of visual system development. Simulations show how newborn
orientation processing and face detection can be specified in terms of
training patterns, and how postnatal learning can extend these
capabilities. The results explain experimental data from laboratory
animals, human newborns, and older infants, and provide concrete
predictions about infant behavior and neural activity for future
experiments. They also suggest that combining a pattern generator with
a learning algorithm is an efficient way to develop a complex adaptive
system.


SELF-ORGANIZATION OF SPATIOTEMPORAL RECEPTIVE FIELDS AND LATERALLY
CONNECTED DIRECTION AND ORIENTATION MAPS
James A. Bednar and Risto Miikkulainen 
To appear in Neurocomputing, 2003 (in press; 8 pages).

http://www.cs.utexas.edu/users/nn/pages/publications/abstracts.html#bednar.cns02

Studies of orientation maps in primary visual cortex (V1) suggest that
lateral connections mediate competition and cooperation between
orientation-selective units, but their role in motion perception has
not been established. Using a self-organizing model of V1 with moving
oriented patterns, we show that (1) afferent weights of each neuron
organize into Gabor-like spatiotemporal receptive fields with ON and
OFF lobes, (2) these receptive fields form realistic joint direction
and orientation maps, and (3) lateral connections develop between
patches with similar orientation and direction preferences. These
results suggest that a single self-organizing system may underlie the
development of orientation selectivity, direction selectivity, and
lateral connectivity.

_______________________________________________________________________________
Software:

LISSOM V4.0: HIERARCHICAL LATERALLY CONNECTED SELF-ORGANIZING MAPS
Available from http://www.topographica.org

James A. Bednar and Jefferson Provost

The LISSOM package contains the C++ and Scheme source code and
examples for training and testing LISSOM-based computational
models. These self-organizing models support detailed simulation of
the development and function of the mammalian visual system, and
include parameter files used to generate the results in the
publications listed above.

Version 4.0 of the simulator provides a graphical user interface (GUI)
for basic tasks, and full batch mode (using a command file) and remote
mode (using a command line prompt) interfaces. The graphs and plots
are available from any of the supported interfaces.

Sample networks are provided for running orientation,
ocular-dominance, motion direction, and face perception
simulations. These existing models can be tested easily with new input
patterns using the GUI, or the command files can be edited to produce
new models based on other training patterns or different network
configurations.

Extensive documentation is included, and is also available via online
help at the command line. The full package is freely available from
our web site, and supports UNIX, Windows, and Mac systems.  For more
details see the tutorial, screenshots, and documentation at
topographica.org.