Typical Sample Selection and Redundancy Reduction for Min-Max Modular Network with GZC Function

摘要

The min-max modular neural network with Gaussian zero-crossing function (M~3-GZC) has locally tuned response characteristic and emergent incremental learning ability, but it suffers from high storage requirement. This paper presents a new algorithm, called Enhanced Threshold Incremental Check (ETIC), which can select representative samples from new training data set and can prune redundant modules in an already trained M~3-GZC network. We perform experiments on an artificial problem and some real-world problems. The results show that our ETIC algorithm reduces the size of the network and the response time while maintaining the generalization performance.