Performance analysis of associative memory using the maximal margin learning

Although the original Hopfield Associative Memory (HAM) causes poor storage capacity and robustness, most of the conventional learning algorithms focus on only improvement in storage capacity or learning speed, and the robustness is hardly considered. We have already proposed a novel learning algorithm based on a maximal margin perspective, which is employed in learning of support vector machines. A HAM learned by the proposed method can much improve the noise reduction effect. In this paper we reveal characteristics of a HAM learned by the proposed method. Several computer simulations show the superior performances and properties of the proposed method to those of the conventional ones.