基于循环多层感知器神经网络的符号逻辑推理系统

Symbol Logic Inference System Based on Recurrent Multi-layer Perceptron Neural Networks

介绍一种用循环多层感知器神经网络实现符号逻辑推理系统的方法。该方法通过让神经网络学习训练样本获取领域规则知识，或者直接将领域规则知识编码于神经网络之中，即用神经网络来表达领域规则知识，然后通过神经网络的循环反馈计算过程来实现任意形式的符号逻辑推理．为研究人类抽象思维（逻辑符号推理）与神经网络形象思维（联接数值计算）之间的关系提供了理论基础。

A method of implementing symbol logic inference system using recurrent multilayer perceptron neural networks is presented in this paper.Domain rule knowledge can be either acquired through learning domain sample set by neural networks or encoded into neural networks directly.Once domain rule knowledge has been stored in a neural networks,the neural networks can be used to implement any symbol logic inference in that domain.It is a theoretical base for studying relations between the abstract thought of human(logic symbol processing) and thinking in images of neural network(linked data calculating).