Abstract
The autocorrelation memory matrix is analyzed with linear algebra and the relations between the memory pattern vectors and the eigenvectors of the memory matrix are investigated. It is elucidated that increase of the memory ratio causes wider distribution of the eigenvalues, which results in loss of memory. The present paper introduces the sign alternating memorization method and shows that this method realizes the seperation of the memory space and the narrower distribution of the eigenvalues. The result of the numerical experiments shows that this method attains greater capacity and wider basin of attraction.
