Browse all publications by topic
Browse all publications by year
- K. Obermayer. Modeling the
Formation of Sensory Representations in the Brain.
.
In Prerational Intelligence, volume 1 of Studies in
Cognitive Systems: Prerational Intelligence, pages 215-232. Kluwer
Academic Publisher, 2000.
Pattern formation by self-organization is a common phenomenon
during brain development. The enormous number of neurons and their
connections makes it impossible for organisms to completely prespecify neural
connectivity patterns within their genomes. Instead organisms seem to specify
processes which then generate the patterns which are observed in the brain.
These processes are supposedly much simpler than the actual patterns, and it
is our hope that there are only few and that they can be cast into a small
set of simple rules. One prominent pattern formation process is the ability
of the brain to form spatial representations of sensory information, which
are commonly called cortical maps and are characterized by the facts (i) that
neurons respond strongly to certain combinations of features, and (ii) that
cells with different response properties are arranged across cortex in
systematic fashion. Well known examples are the multiple topographic
representations of the body surface in the somatosensory areas (Kaas et al.
1979), the various types of acoustic maps in the auditory areas (Suga &
ONeill 1979; Olsen et al. 1989), the visuomotor maps in the superior
colliculus (Tusa et al. 1978; Tusa et al. 1979; Stein & Meredith 1993),
and the retinotopic maps (Dow et al. 1985;), as well as the orientation
(Blasdel & Slaama), ocular dominance (Hubel & Freeman 1979), motion
(Malonek et al. 1993), and stereo maps (Livingstone & Hubel 1988) in the
visual areas of several species. There is also some evidence that spatially
localized representations of information may be involved in cognitive tasks
like language processing in humans (Hart et al. 1985). Modelling studies have
shown that the formation and the plasticity properties of cortical maps can
be described by simple dynamical systems (see, e.g., Swindale 1982; Miller et
al. 1989; Obermayer et al. 1992; Stein & Meredith 1993). It seems that
the phenomenon of cortical maps and their formation provide an example for
our hypothesis: that a small set of amp;quot;primitivesamp;quot; of
information processing and development could account for the variety of
phenomena encountered in brain development and function. I
illustrate this idea by analysing activity dependent processes, and by
investigating their role in cortical map formation. I will mainly concentrate
on the development of cortical maps in the primary visual cortex, in
particular, on the formation of the ocular dominance and orientation maps in
the macaque monkeys area VI. A comparision with experimental
data shows that the developmental rules underlying a simple model based on
Hebbian learning are not sufficient to account for both: the formation of
orientation and ocular dominance. The abstractized version of the Hebbian
approach called the Self-organizing Map generates all the observed phenomena.
This approach is generalized and the formation of various types of cortical
maps is considered.
|
