一类新的指数式联想记忆模型

A Novel Exponential Associative Memory Model

由于Ｋｏｈｏｎｅｎ模型对噪声极端敏感，Ｍｕｒａｋａｍｉ提出一个利用带噪输入优化Ｋｏｈｏｎｅｎ模型的最小平方联想记忆模型（ＬＳＡＭ），大大降低原有模型噪声敏感性。但ＬＳＡＭ的性能依赖于预先给定的输入噪声方差，而这在实际应用中难以达到。本文所提出的一类指数式联想记忆模型，一方面解决了Ｍｕｒａｋａｍｉ所遇到的上述问题，另一方面使原有模型的联想记忆能力得到极大的提高。因此可广泛地应用于模式识别等领域。

Due to the noise sensitivity of Kohonen′s model, Murakami proposes a least square associative memory model (LSAM) by using noise corrupted inputs to optimize Kohonen′s model which greatly reduces the noise sensitivity of the original model. However, the performance of LSAM depends on the predefined variances of input noises, which is unsuitable to application. In this paper, a novel exponential associative memory model is proposed, which overcomes the above shortcomings and greatly raises the associative performance of the former as well. Therefore it can extensively be applied to such fields as pattern recognition.