Flash learning for a multilayer network

The authors propose a learning algorithm, flash learning, that requires only a single presentation of the training set. They introduce a similarity measure for grouping the training examples. The internal representation of a multilayer network, the number of hidden units, and the activation value-space of these units are prestructured before learning based on this group-similarity measure. The flash learning algorithm proceeds to capture the connection weights so as to realize the predetermined activation value-space. The ability to create the prestructured internal representation based on the grouping measure distinguishes flash learning from earlier methods such as back-propagation