联想记忆网络的约束优化学习

CONSTRAINED MINIMIZATION BASED LEARNING FOR ASSOCIATIVE MEMORIES

本文提出了一种联想记忆网络的约束优化学习算法，学习算法是一个全局最小化过程，其初始解保证每个样本是系统的稳定状态，然后逐步增大样本的吸引域，使网络具有优化意义上的最大吸引域．在理论上，我们分析了样本的渐近稳定性和吸引域范围，以及学习算法的收敛性．大量计算机实验结果说明学习算法是行之有效的．

This paper proposes a constrained minimization based learning for associative memory. The learning algorithm is cast into a global minimization problem.It gradually enlarges attraction basins of memory patterns, with a starting point ensuring the storage of these patterns. The asymptotic stability. basin of attraction, and the convergence of the learning method are studied theoretically. Computer simulation results have shown the efficiency of the method.