Version 3.0 of the LISSOM software package for self-organization of
hierarchical laterally connected maps is now available from the UTCS
Neural Networks Research Group website, http://www.cs.utexas.edu/users/nn.
The software has been developed in the LISSOM project of modeling the
mammalian visual system, and is intended to serve as a starting point
for computational studies of the development and function of perceptual
maps in general.
Abstracts of two recent papers from the LISSOM project are also
included below. The first paper shows how LISSOM simulations can be
scaled up to model large cortical areas, obtaining quantitatively
equivalent maps at each size. The second paper uses these techniques
and the LISSOM software to demonstrate how innate face preferences and
later adult face processing may both result from general-purpose
learning and self-organization. Other papers and demos of the LISSOM
software are available at
http://www.cs.utexas.edu/users/nn/pages/research/visualcortex.html.
- - Jim, Amol, and Risto
Software:
- -----------------------------------------------------------------------
LISSOM V3.0: HIERARCHICAL LATERALLY CONNECTED SELF-ORGANIZING MAPS
http://www.cs.utexas.edu/users/nn/pages/software/abstracts.html#lissom
James A. Bednar
The LISSOM V3.0 package contains the C++ source code and examples for
training and testing RF-LISSOM and HLISSOM. These self-organizing
models support detailed simulations of the development and function of
the mammalian visual system. The simulator is designed to have full
functionality even when run in batch mode or remote mode, using a
simple but powerful command file format and online command prompt.
Because of the focus on batch/remote use, it does not have a GUI, but
it does create a wide variety of images for analysis and testing.
Sample command files are provided for running orientation,
ocular-dominance, and face perception simulations using a variety
of network and machine sizes. Extensive documentation is also
included, all of which is also available via online help where
appropriate.
Version 3.0 supports an arbitrary number of maps of various types,
which can be arranged into a hierarchy representing the visual system.
Currently supported map types include input regions (e.g. a Retina),
convolving regions (e.g. ON/OFF cell layers), and RF-LISSOM regions
(with modifiable afferent and lateral connections.) Environmental
input is controlled by a simple but flexible language that allows
arbitrary patterns and natural images to be rendered, scaled, rotated,
combined, etc. This language makes it possible to use LISSOM for many
of your own projects without having to write any new simulator code.
The simulator can also serve as a good starting point for writing a
batch-mode neural-network or related simulator. In particular, it
includes independent and general-purpose routines for image creation
from matrices, PPM format image input and output, gnuplot image
creation, polymorphic datatypes, 2D input drawing, streams of inputs
from different distributions, convolution kernel specification, and
cortical map measurement, as well as many general-purpose support
algorithms and datatypes.
Papers:
- -----------------------------------------------------------------------
SCALING SELF-ORGANIZING MAPS TO MODEL LARGE CORTICAL NETWORKS
Amol Kelkar, James A. Bednar and Risto Miikkulainen
Department of Computer Sciences, The University of Texas at Austin
Technical Report AI-00-285, August 2001.
(Expanded version of CNS*01 paper; 16 pages)
http://www.cs.utexas.edu/users/nn/pages/publications/abstracts.html#kelkar.utcstr01
Self-organizing computational models with specific intracortical
connections can explain many functional features of visual cortex,
such as topographic orientation and ocular dominance maps. However,
due to their computational requirements, it is difficult to use such
detailed models to study large-scale phenomena like object
segmentation and binding, object recognition, tilt illusions, optic
flow, and fovea--periphery interaction. This paper introduces two
techniques that make large simulations practical. First, a set of
general linear scaling equations for the RF-LISSOM self-organizing
model is derived and shown to result in quantitatively equivalent maps
over a wide range of simulation sizes. Second, the equations are
combined into a new growing map method called GLISSOM, which
dramatically reduces the memory and computational requirements of
large self-organizing networks. With GLISSOM it should be possible to
simulate all of human V1 at the single-column level using existing
supercomputers, making detailed computational study of large-scale
phenomena possible.
- -----------------------------------------------------------------------
LEARNING INNATE FACE PREFERENCES
James A. Bednar and Risto Miikkulainen
Department of Computer Sciences, The University of Texas at Austin
Technical Report AI-01-291, November 2001.
(Expanded version of AAAI-00 paper; 28 pages)
http://www.cs.utexas.edu/users/nn/pages/publications/abstracts.html#bednar.utcstr01
Whether humans have a specific, innate perceptual ability to process
faces remains controversial. Studies have found face-selective brain
regions in adults and have shown that even newborns preferentially
attend to face-like stimuli. On this basis researchers have proposed
that there are genetically hard-wired brain regions that specifically
process faces. However, other studies suggest that the face-processing
hardware is general purpose and highly plastic, even at birth. We
propose a solution to this apparent paradox: innate face preferences
may be learned by a general-purpose self-organizing system from
internally generated input patterns, such as those found in PGO waves
during REM sleep. Simulating this process with the HLISSOM model, we
demonstrate that such an architecture constitutes an efficient way to
specify, develop, and maintain functionally appropriate perceptual
organization. This preorganization can account for newborn face
preferences, providing a computational explanation for how genetic
influences interact with experience to construct a complex system.
---------
About this list: http://www.visionscience.com/vsList.html
This archive was generated by hypermail 2b29 : Thu Nov 29 2001 - 21:03:57 GMT