Extraction of logic rules on the basis of robust binary feedforward neural networks

This paper, on the basis of the connection weights of robust binary feedforward neural networks (robust BNNs) being -1, 0 or +1, points out that the extraction of logic rules from robust BNNs is much easier than that from ordinary feedforward neural networks. It also puts forward the point that robust BNNs are a perfect unification of a logic knowledge database, an inference machine and an interpretation machine.