多阈值神经元及其在多值逻辑中的应用

Multi thresholded neuron and it's application in multi valued logic

分析了利用多阈值神经元实现多维空间中多区域非线性划分的原理,以利用神经元实现异或运算为例,阐述了采用多阈值神经元相对于单阈值神经元实现逻辑运算的优点,提出用一个多阈值神经元实现三值逻辑基本运算的方法,从而解决了用多阈值神经网络实现任意三值逻辑函数的问题.在此基础上提出了用一个多阈值神经元实现任意多值函数的方法.此方法简单、规范、有效,可大幅减少神经网络的神经元数量及连线数.

The principle of multi-thresholded neuron (MTN) that implements nonlinear multi-zone partition in multi-dimension space was discussed. The advantage of MTN over single thresholded neuron was shown when applied to implement the XOR operation. An approach of employing MTN for basic operation in ternary logic was proposed to achieve arbitrary ternary functions in neural networks. Based on it, a method was developed for implementation of an arbitrary multi-value functions with single MTN. The proposed method is simple, canonical, effective, and can reduce the number of neurons and connecting lines in neural networks.