Memory Properties of Non-Fully Interconnected Neural Network
R.A. Kosiński and A. Zagórski
Dept. of Applied Physics and Mathematics, Institute of Physics, Warsaw University of Technology, Koszykowa 75, 00-662 Warszawa, Poland
Received: February 4, 1994; revised version: May 13, 1994
Full Text PDF
A model of a neural network consisting of two states neurons with the number N of (symmetric) synaptic connections per neuron treated as a variable was investigated numerically. Hebb's rule was used for storing uncorrelated patterns in the network. A maximal number of such patterns, which can be effectively retrieved by the network and the process of deterioration of the memory, is examined as a function of the number of synaptic connections per neuron. The influence of the number of neurons in the network as well as boundary conditions for the storage capacity of the network are discussed.
DOI: 10.12693/APhysPolA.86.427
PACS numbers: 87.10.+e