|
|
The development of Topographica was supported in part by the
U.S.
National Institutes of Mental Health under
Human Brain Project
grant 1R01-MH66991, and by the U.S.
National Science Foundation under
grant IIS-9811478.
Educational applications of Topographica are supported in part by
the University of Edinburgh Doctoral Training Centre
in Neuroinformatics, with funding from the Engineering and Physical Sciences
Research Council and the Medical Research council through the
Life Sciences Interface. For sample assignments see the
web page
for the course
Computational
Neuroscience of Vision (CNV) offered by the
School of
Informatics of the
University of Edinburgh.
If you use this software in work leading to an academic
publication, please cite this reference:
@Article{bednar:neurocomputing04-sw,
author = "James A. Bednar and Yoonsuck Choe and Judah {De Paula}
and Risto Miikkulainen and Jefferson Provost and Tal
Tversky",
title = "Modeling Cortical Maps with {Topographica}",
journal = "Neurocomputing",
year = 2004,
pages = "1129--1135",
url = "http://nn.cs.utexas.edu/keyword?bednar:neurocomputing04-sw",
}
Many of the ideas in Topographica were developed in conjunction with our
book:
Risto
Miikkulainen, James A. Bednar, Yoonsuck Choe, and Joseph Sirosh.
Computational Maps in the Visual
Cortex.
Springer, Berlin, 2005.
The book has background on cortical maps in general, descriptions of
the various levels of modeling, and a detailed presentation of the
scaling equations that underlie Sheet coordinates
(which are also described in
Bednar et al.
Neuroinformatics, 2:275-302, 2004).
Other useful simulators:
- Neuron and
GENESIS
- Detailed low-level modeling of neurons and small networks. It is possible
to use these simulators for topographic maps, but the computational
requirements are usually extremely high, and typical users simulate
much smaller networks. Note that there are now (3/2007) Python
bindings for Neuron, so it should be practical to wrap a Neuron
simulation into a Topographica Sheet for analysis.
- Catacomb
- Highly graphical Java-based simulator covering numerous levels,
from ion channels to behavioral experiments. Can be used for some
of the same types of models supported by Topographica, but does
not have an explicit focus on topographically organized areas.
- NEST
- NEST (formerly called BLISS) is a general-purpose simulator for
large networks of neurons, but without an explicit focus on
topography. NEST is based on a custom stack-based scripting
language (like RPN calculators or PostScript) that is not nearly
as friendly as Python, and requires much more of the simulation
code to be written in C. On the other hand, NEST does provide
many useful, high-performance primitives, has good parallel
computer support, and can be particularly useful for models that
do not fit Topographica's abstractions closely. NEST now has a
Python interface, which can be used to wrap a spiking NEST
simulation as a Topographica sheet. Thus it is possible to use
the NEST primitives to develop a model that is then controlled
at a higher level by Topographica. Until such an example has
been put into the Topographica release, contact
Jim
for more details if you are interested.
- NCS
- The NCS simulator focuses on large-scale simulation of networks
of spiking neurons, using C/C++ with a custom specification
language rather than an extensible scripting language. Thus it is
likely to be useful primarily for running simulations very similar
to those built by the developers, rather than being fully
extensible as Topographica is.
- iNVT
- iLab Neuromorphic Vision Toolkit is a high-performance
computer-vision oriented C++ toolkit from Koch and Itti with support
for saliency maps for modeling attention. It has a strong focus
on topographically organized regions, but at a high level of
abstraction, and without specific support for learning and
development. As for NCS, it also requires more time-consuming and
less flexible development in C++.
- Emergent
- Formerly called PDP++, Emergent focuses on simulating neural
networks of various types, for either engineering or cognitive
science applications. Although there is support for networks
arranged as maps (e.g. Kohonen SOMs), the interface is designed to
make the influence of individual units clear, which is not
typically useful for analyzing maps. In any case, Emergent has
less emphasis on simulating biological experiments and brain
tissue than does Topographica, instead concentrating on more
abstract systems that perform specific tasks.
- LENS
- Simple, basic artificial neural-network simulator (primarily
abstract backpropagation networks, but also has support for
Kohonen SOM models of topographic maps).
|