An associative memory model based on multiclass classification

The associative memory can be regarded as a multiclass classification problem. Thus, we formulate it as optimization problems to maximize Hamming distances between each prototype and a separate hyperplane. In order to solve them, we propose approximate linear or quadratic programming problems by using L1 or L2 norms. Moreover, we extend the proposed model into a nonlinear model which uses the kernel function. Through some numerical experiments, we verified that the proposed model is effective in the storage capacity and the stability of stored prototypes.