preprint - associative memory in oscillating cortex

[Bill Baird]

 Preprint announcement: available by ftp from neuroprose 


           Learning with Synaptic Nonlinearities in a
          Coupled Oscillator Model of Olfactory Cortex
 
                           Bill Baird

          Depts. Mathematics, and Molecular and Cell Biology,
                 U.C.Berkeley, Berkeley, Ca. 94720



                           Abstract

       A simple network model of olfactory cortex, which assumes only
minimal coupling justified by known anatomy, can be analytically
proven to function as an associative memory for oscillatory patterns.
The network has explicit excitatory neurons with local inhibitory
interneuron feedback that forms a set of nonlinear oscillators coupled
only by long range excitatory connections. Using a local Hebb-like
learning rule for primary and higher order synapses at the ends of the
long range connections, the system can learn to store the kinds of
oscillation amplitude and phase patterns observed in olfactory and
visual cortex. Memory capacity is N/2 oscillatory attractors, N/4
chaotic attractors in an N node network. The network can be truely
self-organizing because a synapse can modify itself according to it's
own pre and postsynaptic activity during stimulation by an input
pattern to be learned. The neurons of the neuron pools modeled here
can be viewed as operating in the linear region of the usual sigmoidal
axonal nonlinearity, and multiple memories are stored instead by the
learned {\em synaptic} nonlinearities.

                         Introduction

        We report recent results of work which seeks to narrow the gap
that exists between physiologically detailed network models of real
vertebrate cortical memory systems and analytically understood
artificial neural networks for associative memory. The secondary
olfactory sensory cortex known as prepyriform cortex is thought to be
one of the clearest cases of a real biological network with
associative memory function.

	Patterns of 40 to 80 Hz oscillation have been observed in the
large scale activity (local field potentials) of olfactory cortex and
visual neocortex, and shown to predict the olfactory and visual
pattern recognition responses of a trained animal. Similar
Observations of 40 Hz oscillation in retina, auditory cortex, motor
cortex and in the EMG have been reported. It thus appears that
cortical computation in general may occur by dynamical interaction of
resonant modes, as has been thought to be the case in the olfactory
system. Given the sensitivity of neurons to the location and arrival
times of dendritic input, the sucessive volleys of pulses that are
generated by the oscillation of a neural net may be ideal
for the formation and reliable longe range transmission of the
collective activity of one cortical area to another.

	The oscillation can serve a macroscopic clocking function and
entrain or ``bind" the relevant microscopic activity of disparate
cortical regions into a well defined phase coherent collective state
or ``gestalt". This can overide irrelevant microscopic activity and help
produce coordinated motor output. If this view is correct, then
oscillatory network modules form the actual cortical substrate of the
diverse sensory, motor, and cognitive operations now studied in static
networks. It must ultimately be shown how those functions can be
accomplished with oscillatory dynamics.


ftp proceedure:

	unix> ftp cheops.cis.ohio-state.edu
	Name: anonymous
	Password: neuron
	ftp> cd pub/neuroprose
	ftp> binary
	ftp> get baird.oscmem.ps.Z
	ftp> quit
	unix> uncompress baird.oscmem.ps.Z
	unix> lpr -P(your postscript printer) baird.oscmem.ps
 
	For background papers, send e-mail to baird at icsi.berkeley.edu,
giving paper or e-mail address for Tex or Postscript output.