Output details

Computational capabilities of multilayer committee machines

<22> This article explored for the first time a special kind of feed-forward neural network: the Ultrametric committee machine. These networks represent the next step in architectural complexity after the perceptron and can be used to separate continuous binary classes with very convoluted boundaries. In particular the paper analysed the measure of difficulty associated to such networks when they act as teachers in an on-line learning scenario. This work is the result of a collaboration with Prof. Leonardo Franco, Universidad de Malaga, Spain.