Connectionists: Modelling software + major V1 modelling review paper
James A. Bednar
jbednar at inf.ed.ac.uk
Mon Nov 26 06:13:24 EST 2012
I am pleased to announce the release of a set of software packages for
modelling neural regions and systems, along with a new review paper
that ties together work from dozens of projects with collaborators
using this software to model the primary visual cortex.
The review paper shows how these separate projects add up into what is
hoped to be the most complete model of the development of functional
properties of V1 neurons to date:
James A. Bednar.
Building a mechanistic model of the development and function of the primary visual cortex.
Journal of Physiology-Paris, 106:194-211, 2012.
http://dx.doi.org/10.1016/j.jphysparis.2011.12.001
http://homepages.inf.ed.ac.uk/jbednar/papers/bednar.jpp12.pdf
The freely available, cross-platform (Linux, Mac, Windows) Python
software packages include:
Param/ParamTk: Handling parameters used in scientific programs
ImaGen: Generating input patterns and connection patterns
Topographica: Modelling interconnected neural regions
More details about each package and the review paper are included below.
Links to 31 publications using this software so far are available here:
http://topographica.org/Home/pubs.html
Thanks to the 30+ collaborators who have contributed to the research
and software development reported here, each cited in the review
paper or listed on topographica.org. We would very much appreciate
any feedback, suggestions, or ideas for future collaboration.
Jim
James A. Bednar, Ph.D.
Senior Lecturer, University of Edinburgh
Director, Computational Systems Neuroscience Group
http://homepages.inf.ed.ac.uk/jbednar/research.html
Director, Edinburgh Doctoral Training Centre in
Neuroinformatics and Computational Neuroscience
http://anc.ed.ac.uk/dtc
_______________________________________________________________________________
OPEN-SOURCE SOFTWARE PACKAGES
Param 1.0 released 7/2012 (http://ioam.github.com/param/):
The Param library makes it simple to add support for Parameters;
Param has no dependencies and is very lightweight, so that it can
be used with any Python program. A Parameter is a special type of
Python attribute extended to have features such as type and range
checking, dynamically generated values, documentation strings,
default values, etc., each of which is inherited from parent
classes if not specified in a subclass. Parameters are extremely
useful for writing scientific software, making it clear which
values are intended to be changed in practice and avoiding
potentially dangerous user errors.
ParamTk 0.8 released 7/2012 (http://ioam.github.com/paramtk/):
Optional extension to Param that provides a GUI for editing
parameter values for your objects without requiring any
GUI-specific coding.
ImaGen 1.0 released 7/2012 (http://ioam.github.com/imagen/):
Provies comprehensive support for creating resolution-independent
spatial pattern distributions. ImaGen consists of a large library
of primarily two-dimensional patterns, including mathematical
functions, geometric primitives, images read from files, and many
ways to combine or select from any other patterns. These patterns
can be used in any Python program that needs configurable patterns
or a series of patterns, with only a small amount of user-level
code to specify or use each pattern.
Topographica 0.9.8 released 11/2012 (http://topographica.org/):
Topographica allows researchers to set up models of complete neural
regions and systems relatively easily, because it takes care of a
lot of the otherwise-painful details of spatial coordinate systems,
mapping between brain regions and between layers in the same
region, scaling between different sampling densities, having
spatially restricted patterns of connectivity, specifying input and
weight patterns (via ImaGen), and measuring tuning curves,
receptive fields, and maps. Topographica is a general-purpose
object-oriented event-driven simulator that provides extensive
flexibility, with families of parameterized objects that can be
customized and adapted for new modelling projects. Current models
primarily focus on the visual system, but they are implemented
using generic primitives that have also been used for somatosensory
and auditory cortex modelling, as well as subcortical and
motor-output models.
_______________________________________________________________________________
The review paper (citation below) describes the GCAL model, which
shows how a relatively small number of simple biological mechanisms,
based on Hebbian learning and homeostatic plasticity, can lead an
unorganized neural region to develop:
- Receptive fields selective for orientation, ocular dominance,
motion direction, spatial frequency, disparity, and color
- Preferences for each of these organized into realistic topographic
maps
- Lateral connections between neurons that reflect the structure of
the maps, as found experimentally
- Both simple and complex cells
The resulting neurons exhibit:
- Realistic surround modulation effects, including their
diversity, caused by interactions between these neurons
- Contrast-gain control and contrast-invariant tuning, ensuring that
they retain selectivity robustly
- Long-term and short-term plasticity (e.g. aftereffects),
emerging from mechanisms originally implemented for development
These properties each arise from an initially undifferentiated
cortical region model, suggesting that the mechanisms involved will
also explain a large variety of cortical phenomena across different
areas and modalities.
@Article{bednar:jpp12,
title = "Building a Mechanistic Model of the Development and
Function of the Primary Visual Cortex",
author = "James A. Bednar",
journal = "Journal of Physiology - Paris",
year = 2012,
volume = 106,
pages = "194--211",
url = "http://dx.doi.org/10.1016/j.jphysparis.2011.12.001",
urlalt = "http://homepages.inf.ed.ac.uk/jbednar/papers/bednar.jpp12.pdf",
abstract = "Researchers have used a very wide range of different
experimental and theoretical approaches to help
understand mammalian visual systems. These
approaches tend to have quite different assumptions,
strengths, and weaknesses. Computational models of
the visual cortex, in particular, have typically
implemented either a proposed circuit for part of
the visual cortex of the adult, assuming a very
specific wiring pattern based on findings from
adults, or else attempted to explain the long-term
development of a visual cortex region from an
initially undifferentiated starting point. Previous
models of adult V1 have been able to account for
many of the measured properties of V1 neurons, while
not explaining how these properties arise or why
neurons have those properties in particular.
Previous developmental models have been able to
reproduce the overall organization of specific
feature maps in V1, such as orientation maps, but
the neurons in the simulated maps behave quite
unlike real V1 neurons, and in many cases are not
even testable on actual visual stimuli because the
developmental models are so abstract.
In this review of results from a large set of models
developed from shared principles and a set of
underlying software components, I show how these
models represent a single, consistent explanation
for a wide body of experimental evidence, and form a
compact hypothesis for much of the development and
behavior of neurons in the visual cortex. The
models are the first developmental models with
wiring consistent with V1, the first to have
realistic behavior with respect to visual contrast,
the first to include all of the demonstrated visual
feature dimensions, and the first to have wiring
compatible with anatomical results. The goal is to
have a comprehensive explanation for why V1 is wired
as it is in the adult, and how that circuitry leads
to the observed behavior of the neurons during
visual tasks.",
}
--
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.
More information about the Connectionists
mailing list