Neural network learning time: effects of network and training setsize

摘要

The learning time for two-layer backpropagation networks isexamined in the context of learning Boolean logic equations fromexamples. In particular, the relationship between the number of inputs,hidden units, and training set vectors and the learning time isinvestigated. The networks, the training algorithm, and the tasks aredescribed. The parameter variations and the set of simulations performedare detailed. Training and test set generation are discussed, and thesimulation results are summarized. Network performance is evaluated, andan alternate training methodology that may remedy problems inherent tothe backpropagation training method is presented