Investigation of generalized Hopfield model by statistical physics methods

Boris V. Kryzhanovsky, Leonid B. Litinskii
2010 The 2010 International Joint Conference on Neural Networks (IJCNN)  
The proposed generalization of the Hopfield model consists in assigning different weight coefficients to input patterns that are used to construct the Hebb connection matrix. Application of statistical physics methods to such a system gives unexpected results even for the simplest variant of differences of weight coefficients. Namely, there are unusual behaviors of the critical value of the load parameter and of the overlap of the state with the pattern.
doi:10.1109/ijcnn.2010.5596872 dblp:conf/ijcnn/KryzhanovskyL10 fatcat:bruikt346jasxbagztlxv5in2u