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。 Since Kohonen model is extremely sensitive to noise, Murakami proposed a Least Squares Associative Memory Model (LSAM) that uses the noisy input to optimize the Kohonen model, which greatly reduces the noise sensitivity of the original model. However, the performance of LSAM depends on the given input noise variance, which is hard to achieve in practical applications. The exponential associative memory model proposed in this paper solves the above problems encountered by Murakami on the one hand, and greatly improves the associative memory ability of the original model on the other hand. It can be widely used in the field of pattern recognition.