Temporal Correlation Between Stimulus And Response In Spiking Neural Networks

\setlength{\parindent}{10ex}
It is well known that temporal processing on time scales of tens and
hundreds of milliseconds plays an important role in simple and complex
sensory problems in the brain, such as, motion discrimination, and
speech recognition. Recent studies showed that cultured cortical
networks can be shaped by the history of an external stimulus by
reflecting the temporal patterns of these stimuli in the network
dynamics, suggesting that cortical networks are capable of learning
the stimulus time scales. Although one might know that synaptic
plasticity may be a key process in this phenomenon, the detailed
process underlying it still demands for an explanation.

\setlength{\parindent}{10ex}
In this work we are concerned in exploring, mathematically and
computationally, by means of a bottow-up model , the possible
mechanisms that take place in these time events. For that we make use
of a conductance-based integrate-and-fire model with a
spike-timing-dependent-plasticity (STDP) rule. We train the network
with a special-temporal pattern, similar to previous experimental
works, and search for the network topology and plasticity effects. We
use networks with three different topological connections: random, regular
square lattices and small world, with different connectivity
and connection probabilities, and change gradually the parameters of
the STDP model in order to control the plasticity action.

\setlength{\parindent}{10ex}
We have verified that the minimum frequencies that allow a
self-sustained activity is closely dependent of network size. Namely
small networks demand higher frequencies than bigger ones, and usually
asynchronous irregular (AI) states are only found in large networks.
Assuming that this kind of state is important for modeling, since it
is observed in an awake cortex, we also added poissonian processes in
a fraction of neurons randomly chosen, mimicking a network of bigger
size. This produced the expected AI behavior and turned possible the use of
realistic synaptic weights and the acquisition of firing
rates of $\sim$ 10 Hz. This work is still in progress, and the
step we are implementing now is the fine tuning of the stimulus-response
system. This will allow us to evaluate the time scales for which the network
is able to respond to the stimulus.