|
|
Our group
partecipates in the Italian national project of INFN named T061
titled
"
Biological Application of theoretical physics methods"
The activity of our group inside TO61
in the last
few years focus on the use of theoretical
physics tools, and in particular statistical mechanics and stochastic models, to
modelling a system as complex as the brain.
The goal is to investigate processing and
imprinting of information in the brain, focusing on cortical
dynamics, plasticity and oscillations.
Cortical areas indeed play a key role in important functions like
those related to the memory.
Recently the amount of experimental data available to
computational neuroscience community grows
steadily, due to the development of new
experimental technique, such as multiunit electrical recordings
which enable the activity of large
populations of neurons to be followed
simultaneously. The data obtained with these modern techniques
allows a degree of comparison with modeling
results that so far was not possible. At
the same time new and stronger theoretical ideas need to be
developed for the modeling, borrowing ideas
and tools from different fields of science,
such as theoretical physics.
The effects of noise on the dynamics of a simple stochastic model
of the cortex has been investigated by the
group [M.Marinaro S.Scarpetta Phys.Rev.E
70, 2004] and compared with some multiunit electrical recordings
of cortical cultures available in
literature. A model of the lamprey neural
spinal cord pattern generator has been investigated by us [Z.Li,
A.Lewis, S.Scarpetta Pys.Rev.Lett. 2004] in collaboration
with partners of the UCL University.
A learning rule inspired to the recently observed
Spike-Timing-Dependent-Plasticity (STDP) has been
introduced and analyzed by the groupin collaboration with John
Hertz (Nordita, DK) and Z.Li (UCL, London)
[Scarpetta et al. Neural Computation 2002, Marinaro et al.
Mathematical Biosciences 2006, M. Yoshioka
S.Scarpetta et al PhysRevE in press 2007].
Oscillations are ubiquitos in the brain. Role of cortical
oscillations and STDP in the hippocampus's theta phase precession
phenomena have been analyzed by the group [Marinaro et al
Hippocampus 2005].
Recently [M. Yoshioka S.Scarpetta M Marinaro PhysRevE in press
2007] we studied spatio-temporal learning in analog neural
networks, assuming the spike-timing-dependent synaptic plasticity
(STDP) in the learning rule. When encoding the finite number of
periodic spatiotemporal patterns, we derive
the order parameter dynamics of the networks. This dynamics
clarifies that the analog neural networks
with the STDP-based learning rule act as associative memory for
the periodic spatio-temporal patterns. The retrieval solution of
the order parameter dynamics elucidates that the phase of the
Fourier transform of the STDP time window determines the retrieval
frequency, while the time average of the STDP time window
crucially affects the storage capacity. The stability analysis of
the retrieval solution indicates that under a certain condition,
stable retrieval state with a single encoded pattern becomes
unstable with multiple encoded patterns even when the retrieval
state is independent of the pattern number.
To examine the wide applicability of the STDP-based learning rule,
we also investigate learning of spatiotemporal Poisson patterns.
Our numerical simulations demonstrate that the Poisson patterns
are memorized successfully both in analog neural networks and
spiking neural networks.
Our
group organize each year the International School on Neural Nets
"E.R.Caianiello"
This year there
is the 12th Course
: " Dynamic Brain"
LIST OF PUBBLICATIONS
|
|
