Extracting Rules from the Aggregate Feedforward Neural Network

摘要

The Aggregate Feedforward Neural Network (AFFNN) is a connectionist model that learns the relationships leading to all K input attributes using a single neural network. This unified network behaves as if it is an aggregation of K distinct networks, each trained for a specific target. This paper describes the creation of the AFFNN using one of the well-known Three MONK's Problems, and then demonstrates the use of a decompositional rule extraction approach for obtaining and expressing the knowledge learned by the AFFNN.