Random brains

摘要: In this paper, we introduce and evaluate a novel method, called random brains, for producing neural network ensembles. The suggested which is heavily inspired by the forest technique, produces diversity implicitly using bootstrap training randomized architectures. More specifically, each base classifier multilayer perceptron, number of randomly selected links between input layer hidden are removed prior to training, thus resulting in potentially weaker but more diverse classifiers. experimental results on 20 UCI data sets show that brains obtained significantly higher accuracy AUC, compared standard bagging similar networks not utilizing analysis shows main reason increased ensemble performance ability produce effective diversity, as indicated increase difficulty measure.