The rule-extraction through the preimage analysis
Rua‐Huan Tsaih, Yat‐wah Wan, Shin‐Ying Huang · 2008
This study reveals the properties of the input/output relationship for a real-valued single-hidden layer feed-forward neural network (SLFN) with the tanh activation function on all hidden-layer nodes and the linear activation function on output node. Specifically, the rule-extraction of the SLFN is done through mathematically analyzing its preimage, which is the set of input values for a given output value.