A Spectral Domain Associative Memory with Improved Recall

A new associative memory model is proposed on the basis of a nonlinear transformation in the Fourier domain of the data. The Moore-Penrose pseudoinverse is used to compute the optimal leastsquares solution. Computer simulations, using onedimensional speech and two-dimensional images, are presented. Comparison of the new model with the classical optimal linear associative memory and an optimal nonlinear (polynomial) memory is presented. 1 Introduction Associative recall (memory) may be understood as an operation or transformation with a set of input signals or other items considered as keys, and some sort of outcome which constitutes the recall. The most characteristic property of associative recall is the following: if the totality of the input signal, is stored as such in a memory, it is retrievable by a part of the key. A recall is autoassociative if the pattern is retrievable on a basis a fragment of the key. In the heteroassociative mode of operation, an outcome which structural..