Understanding Neuronal Coding
Dr Chris Christodoulou
The problem of understanding neuronal coding (i.e., how the brain encodes/decodes
and transmits information) and what determines highly variable firing in
real neurons ranks today amongst the most important fundamental issues
in neuroinformatics/computational neuroscience. This is because a solution
would provide the basis for the analytical evaluation of the brain's information
processing capability and would give us a further insight as to those aspects
which are essential to its functional organisation. In particular (in collaboration
with researchers from the Univ of Plymouth and King's College, Univ of
London), using leaky integrator type models we have identified partial
somatic reset as one of the mechanisms for producing irregular firing (paper
in Neural Computation, 1997) and for controlling the gain of the neuron.
In the same paper we have also shown that temporal integration of random
input spikes and input current fluctuation detection can coexist and cooperate
to cause highly irregular firing. In addition, we have demonstrated more
recently (papers in Biosystems, 2000 and in Neural Networks, 2002), that
with 80% of inhibition on concurrent excitation, firing becomes nearly
irregular. Furthermore, by examining and comparing the biological mechanisms
of models used to produce irregular firing, we found that the most likely
candidate for doing that is the partial somatic reset one (paper in Neurocomputing,
2001). We have also demonstrated (see paper in Neurocomputing, 2002), that
when an integrate-and-fire (I&F) neuron model (which is the simplest
spiking neuron model) with random synaptic input is equipped with the partial
somatic reset mechanism, its firing is only very weakly dependent on the
level of inhibitory input. This supports (in contrast to previous suggestions)
that the I&F neuron can replace more biophysical models in stochastic
neuronal modelling.
Our work above and the work of others have identified most of the determinants
of irregular neural firing. There are still though some very fundamental
questions that we are currently working on:
1. How do we ascertain whether a particular firing pattern does indeed
represent a `code`, i.e., one with a functional role in representation,
transmission and processing of information and
2. What type of code it is (rate code, temporal code, a combination
of the two or another type) and what is its relation, if any, with the
input neuronal parameters and signals that produced it.
Publications
-
Christodoulou, C. (2002). On the firing variability of Integrate-and-Fire
neurons with partial reset in the presence of inhibition. Neurocomputing,
44-46 , 81-84.
-
Christodoulou, C. and Bugmann, G. (2001). Coefficient of Variation
(CV) vs Mean Interspike Interval (ISI) curves: what do they tell us about
the brain? Neurocomputing, 38-40, 1141-1149.
-
Christodoulou, C. and Bugmann, G. (2000). Near-Poisson-Type Firing
Produced by Concurrent Excitation and Inhibition. Biosystems
, 58, 41-48.
-
Bugmann, G., Christodoulou, C. and Taylor, J. G. (1997). Role of
the Temporal Integration and Fluctuation Detection in the Highly Irregular
Firing of a Leaky Integrator Neuron with Partial Reset. Neural Computation
, 9, 5, 985-1000.
This paper also appeared in:
-
Bugmann, G., Christodoulou, C. & Taylor, J. G. (1999). Role
of temporal integration and fluctuation detection in the highly irregular
firing of a Leaky Integrator neuron with partial reset. In: Neural Codes
and Distributed Representations: Foundations of Neural Computation. L.
Abbott and T. J. Sejnowski (eds.), MIT Press (ISBN 0-262- 51100-2), 171-186.
(Note: This book "collects, by topic, the most significant papers
that have appeared in the journal Neural Computation over the past
nine years"
Revised on 9th September 2002
webmaster@dcs.bbk.ac.uk