Friday, April 27, 2001
Netherlands Institute for Brain Research
Meibergdreef 33, 1105 AZ Amsterdam,
The Netherlands
map
Organization:
Jaap van Pelt
The objective of the present Seminar Computational Neuroscience is
to discuss recent developments in Computational Neuroscience and to
bring together scientists within The Netherlands working in or
interested to explore research possibilities in this
multidisciplinary field of research.
Location:
Netherlands Institute for Brain Research
Meibergdreef 33, 1105 AZ Amsterdam
The Netherlands
Tel: +31 20 5665500; Fax: +31 20 6961006
map
Modeling neuronal morphogenesis and network formation
Arjen van Ooyen, Jaap van Pelt
Neurons communicate through
electrical and chemical signals, traveling along their
dendritic and axonal arborisations.
Neurons show an enormous variety in shapes which
is believed to reflect their functional specialisation.
Neurons attain their shapes as the result of a developmental process
in which intracellular mechanisms and interactions with local
environments are operating in concert. Many of these mechanisms
are modulated by bioelectric activity. It is believed that these
modulatory influences play a key role in the structural and functional
tuning of the nervous system.
Mathematical modeling and computational approaches are
essential to obtain understanding of these dynamical processes.
Examples will be discussed, concentrating on
growth of dendritic arborisations, competitive phenomena during
neurite
outgrowth and formation of nerve connections, and
implications of activity-dependent processes for neural
development and network formation.
Modeling axonal guidance
Boris Lastdrager.
During the development of the nervous system neurons are connected to
their targets by axons which grow from the neurons toward the targets,
a process which is supposed to be governed by the secretion and
detection of chemoattractants and chemorepellants. During the growth
the axons are seen to form a bundle in the initial growth phase which
then grows towards the target area where the axons debundle and attach
to individual targets. We will discuss a simple model in which axons
interact by secreting chemicals from their growth cones which are
diffused through the extracellular space and sensed by other growth
cones. With this simple model we can reproduce the processes of
bundling, debundling and attachment which lends credibility to
theories that describe axonal growth through similar simple models. In
the implementation of the model we discovered a numerical pitfall that
relates to the implementation of the source terms that can easily
render a code useless. In our implementation we have managed to avoid
this problem. It turns out that the model we consider is very
sensitive to parameters, i.e., a slight change in a parameter, e.g. a
diffusion constant, can prevent bundling, debundling or attachment.
The parameter values for which bundling, debundling and attachment
occur are however within ranges corresponding with experimental
measurements.
Mathematical model for electrical bursting and Ca2+ oscillations in`
neuroendocrine cell
Niels Cornelisse
Nonlinearity is essential for the computational function of the brain
and occurs at various levels of the brain. At the cellular level the
signaling between neurons is highly nonlinear due to the complex
interactions between neurotransmitters, receptors, second messengers
and ion channels that are involved. We study cellular signaling in a
neuroendocrine cell of the amphibian Xenopus laevis that has a clear
cellular response (hormone secretion) to neuronal input. A
mathematical model has been developed to describe the specific
features of the cellular signaling in these cells (electrical
bursting, Ca2+ oscillations). The model is an extended Hodgkin-Huxley
type model partly based on experimental data and is analysed with the
use of nonlinear system analyses, phase space analyses and bifurcation
theory.
On the computational implications of dynamical synapses
Bert Kappen
Recent experiments in pyramidal cells show that the amplitude of the
postsynaptic membrane potential (PSP) depends on presynaptic activity.
Although the details of this phenomenon depends on the type or synapse
and the brain area, a common trend is that synaptic strenght
decreases as a result of presynaptic activity with a characteristic
time scale of about 10 msec. Recovery occurs in the order of
seconds. These synaptic mechanisms can have quite profound functional
consequences on the behaviour of neural networks. For instance,
attractors in recurrent neural networks will become metastable states
due to the changing connectivity pattern. In this presentation I will
present both numerical and analytical results of this new phenomenon.
How the brain might perform motor tasks: From backpropagation and
Boltzmann machines to biologically realizable machines
Willem van Leeuwen
As is well-known, backpropagation and Boltzmann machines yield ways
to adapt the weights (synapses) of a neural net in such a way that
desired input-output relations (for instance motor tasks)
can be realized
by the net. However, since the existing algorithms, like
`backpropagation'
and `Boltzmann machines' require an outside observer and outside
computer, these
algorithms do not help us to understand the way in which the actual
brain might work.
Recently, the neuroscientists D.R. Chialvo and P. Bak of the Divison
of Neural Systems in Copenhagen suggested a hormone controlled way to
adapt the synapses. This control mechanism might be realized by an
actual brain to adapt itself in such a way
that it can perform the desired input-output relations. By combining
the suggestion of Chialvo and Bak at the one hand and the recipe
called
Hebbian learning at the other hand, the theoretical physicists
Reinier Bosman and Bastian Wemmenhove of the Institute of Theoretical
Physics in Amsterdam could make the Danish idea not only more
realistic
biologically, but also more effective.
Modeling transitions in cognitive development with
bifurcations in neural networks
Maartje Raijmakers
In the field of cognitive science, artificial neural network
simulation usually is carried out with configurations that are, from a
developmental-biological point of view, in a mature state. That is,
the
contours of neural field organization are given and simulation usually
pertains to the adaptation of interconnections within and between the
given fields. Recently, constructive neural networks of development
model maturation by the addition of structure during and in addition
to learning (e.g., Elman, 1993; Quartz, 1994). In these cases, more
inspired by computation theory than by biology, resources, such as
hidden units are added to the network as a function of learning. Our
future objective is to model the maturation process of a cognitively
applied neural network by ordinary differential equations that
describe the evolution of real valued parameters as a function of
activation. As will be shown, in the plane of these parameters local
bifurcations of activation dynamics occur. That is, parameter changes
trigger qualitative changes in network dynamics and, as a consequence,
qualitative changes in the cognitive network behavior as well. The
latter induction of qualitative new cognitive behavior without adding
resources is a consistent interpretation of epigenetic theories of
cognitive development, like Piaget's stage theory (Molenaar, 1986;
Raijmakers, 1997).
Detection of connections among image regions by the visual brain
Pieter R. Roelfsema
Connectedness is one of the most important grouping criteria that
allow the visual system to segregate objects from each other and from
the background. Algorithms for the detection of connectedness will be
reviewed from a neurophysiological and psychophysiological point of
view. It will be suggested that connectedness detection by a
feedforward network is physiologically implausible. Instead, I present
evidence that visual cortical neurons label connected image regions
serially, by exhibiting an enhanced firing rate. Psychophysical
results imply that connectedness detection requires visual attention.
Niels Cornelisse
Dept. Medical & Biophysics, Dept. Cellular Animal Physiology,
Nijmegen Institute for Neurosciences
University of Nijmegen, Geert Grooteplein 21,
PO Box 9101, 6500 HB Nijmegen, The Netherlands
Tel: +31-24-3614235; Fax: +31-24-3541435
Email: cornelis@sci.kun.nl
Bert Kappen
SNN, University of Nijmegen
Tel: +31-24-3614241; Fax: +31-24-3541435
Email: bert@mbfys.kun.nl
http://www.mbfys.kun.nl/~bert
Boris Lastdrager
CWI
Kruislaan 413, P.O. Box 94079, 1090 GB Amsterdam, The Netherlands
Phone: +31-20-5924077; Fax: +31-20-5924199
Email: Boris.Lastdrager@cwi.nl
Willem van Leeuwen
Institute for Theoretical Physics, University of Amsterdam
Valckenierstraat 65, 1018 XE Amsterdam, The Netherlands
Tel: +31 20 525 5773 (5747); Fax: +31 20 525 5788
E-mail: leeuwen@science.uva.nl
http://soliton.wins.uva.nl/~leeuwen/
Arjen van Ooyen
Netherlands Institute for Brain Research
Meibergdreef 33, 1105 AZ Amsterdam, The Netherlands
Tel: +31-20-5665483; Fax: +31-20-6961006
Email: a.van.ooyen@nih.knaw.nl
http://www.anc.ed.ac.uk/~arjen
Jaap van Pelt
Netherlands Institute for Brain Research
Meibergdreef 33, 1105 AZ Amsterdam, The Netherlands
Tel: +31-20-5665481; Fax: +31-20-6961006
Email: j.van.pelt@nih.knaw.nl
Maartje Raijmakers
Department of Psychology, University of Amsterdam
Roeterstraat 15, 1018 WB Amsterdam, The Netherlands
Tel. +31-20-5256826; Fax. +31-20-6390279
Email: op_raijmakers@macmail.psy.uva.nl
Pieter Roelfsema
IOI, Afdeling Visuele Systeem Analyse, Academisch Medisch Centrum
Meibergdreef 15, 1105 AZ Amsterdam
Tel. +31-20-5665603
Email: p.roelfsema@ioi.knaw.nl
Erik de Schutter
Theoretical Neurobiology,
Born Bunge Stichting, Universitaire Instelling Antwerpen UIA
Universiteitsplein 1, B2610 Antwerpen, Belgium
Email: erik@bbf.uia.ac.be
Wytse Wadman
Institute for Neurobiology, University of Amsterdam
Kruislaan 320, 1098 SM Amsterdam, The Netherlands
Tel: +31-20-5257641; Fax: +31-20-5257709
Email: wadman@bio.uva.nl