We present an attractor neural network which can associatively retrieve a variety of activity patterns encoded in the synaptic matrix between the excitatory neurons. The neurons are characterized by an absolute refractory period of 2 ms and would at saturation emit spikes at a rate of 500 s-1, yet the collective operation of the network allows stable retrieval performance at rates as low as 20-25 s-1. The network is presented as a model of increasingly realistic neurons assembled in a network with increasingly realistic output structures, on which a variety of experiments can be carried out. The types of features included are: continuous dynamics of the membrane potential except at spike emission; differentiation of excitatory and inhibitory operation; relative refractory period, due to post-spike hyperpolarization; membrane potential decay constants; uniform or random spike transmission delays; inhibition by hyperpolarization or by shunting; short and persistent stimuli, represented as synaptic currents into a fraction of the neurons in a pattern and into some other neurons; noise is modelled as random continuous afferent correlated in time due to the gradual decay of the membrane potential. Features studied are the control of the activity rates during retrieval by the noise level; the robustness of orderly retrieval to the introduction of spike transmission delays, uniform and random; the appearance of oscillations during retrieval due to global synaptic modification with uniform and random delays; the effect on retrieval of persistent weak stimulus; storage capacity and its relation to persistent stimuli and delay type. No analysis is attempted. Instead an effort has been made to provide probes into the operation of the network. Those consist of the possibility of measuring on line intra-neural membrane potentials of selected neurons; obtaining spike rasters of selected groups of neurons and a moving bin average activity rate in each of the three groups of neurons: excitatory in the pattern recalled, excitatory out of the pattern and inhibitory.
|Number of pages||25|
|Journal||Network: Computation in Neural Systems|
|State||Published - 1990|
Bibliographical noteFunding Information:
DJA would like to acknowledge discussions with Drs Larry Abbot, A Nowak, J Buh-mann and M Tsodyks. We are also indebted to Dr J Buhmann for providing us with his simulation programs. The work of DJA has been supported by a grant from the US-Israel Bi-National Science Foundation, to which we also owe the SUN network on which these extensive simulations were carried out. ME is grateful to the Lady Davis Fund for a fellowship which allowed his stay in Jerusalem during this project.