Storage capacity of neural networks storing spatially correlated patterns
Mathias Schlüter, Friedrich Wagner · Physical review. E, Statistical physics, plasmas, fluids, and related interdisciplinary topics · 1994
The storage capacity of Ising spin networks storing spatially correlated patterns with explicit learning rules is investigated. The correlations are introduced by using equilibrium configurations of the Ising model at temperature 1/\ensuremath{\beta}. Using the Hebb rule, the storage capacity decreases strongly with increasing \ensuremath{\beta}. This can be avoided by a modification of the Hebb rule which includes the inverse equilibrium correlation matrix of the Ising model. The theoretically derived expressions for the storage capacity are in good agreement with the numerical simulation. In order to demonstrate the working mechanism of the alternative learning rule the storage of simple linear patterns is discussed in detail.