Estimates of optimal storage conditions in neural network memories based on random matrix theory
Wojciech Tarkowski, Maciej Lewenstein · Journal of Physics A Mathematical and General · 1992
The authors formulate a method for estimating the critical conditions for storage of sets of data in neural network memory. This variational method is based on random matrix theory and calculating the average spectrum of a matrix, whose elements are given by overlaps of the stored patterns. Several generic cases of random overlap matrices are considered. They investigate the cases of simply uncorrelated random patterns and 'spatially' and 'semantically' correlated ones. They obtain bounds of the critical curve in the control parameters space, which determine the stability of the stored data sets.