基于结构学习和迭代自映射的自联想记忆模型

A Homogenous Associative Memory Model Based on Structure Learning and Iterative Self-Mapping

传统的人工神经元网络连接结构是固定的,是对权值的学习.提出一种基于生理神经元特征的人工神经元模型,并在以此为单元构成的用于实现自联想记忆的神经网络上进行对结构的学习.学习算法以设定神经元的输入/输出感受野、调整突触和轴突末梢的连接、并行的自投影迭代为特征.给出了此网络模型的矩阵描述和实验结果.

The connections of traditional ANN (artificial neural network) are fixed, and the learning algorithm is to change the weight of every connection. In this paper, a new model is presented which is based on the characteristic of physiological neuron, and as a unit, a homogenous associative memory neural network was built, the learning algorithm was performed to change the structure of the neural network. The characteristic of this algorithm sets the input and output field of a neuron, adjusts the connection between synapse and axon and parallel iterative self-mapping. The matrix model of the network and the experiment results are also presented in this paper.