CAUSAL RELATION RULE EXTRACTING METHOD BY NEURAL NETWORK

摘要

PURPOSE:To extract a clear causal relation rule without redundancy by finding Boolean algebraic expressions about each local connection after changing a neural network into a skeleton network and reducing a connection number to the necessary minimum. CONSTITUTION:On the parameter tuning of an original network 1, a skeleton network 4 to be reduced a connection distributing degree adopting a skeleton studying method is generated. And with regard to each local connection when making the connection of one output node and all input nodes related to this in this skeleton network 4 as one unit, the lattice structures 5A-5C of Boolean product concerning the variable of the input node is prepared, and a product sum Boolean algebraic expression as the causal relation rule in the local connection is obtained by searching the true minimum vertical angle of the lattice structure 5A-5C. Thus, it is possible to extract the clear causal relation rule without the redundancy.