Rule extraction in trained feedforward neural networks with first-order logic

Abstract

The need for neural-symbolic integration becomes evident as more complex problems are addressed, and that go beyond limited domain tasks such as classification. The search methods for extracting rules from neural networks work by sending input data combinations that activate a set of neurons. By properly ordering the input weights of a neuron, it is possible to narrow the search space. Based on this observation, this paper aims to present a method to extract the rule pattern learned by a feedforward trained neural network, analyze its properties and explain these patterns through the use of first-order logic (FOL).