M. Rattray et D. Saad, GLOBALLY OPTIMAL ONLINE LEARNING RULES FOR MULTILAYER NEURAL NETWORKS, Journal of physics. A, mathematical and general, 30(22), 1997, pp. 771-776
We present a method for determining the globally optimal on-line learn
ing rule for a soft committee machine under a statistical mechanics fr
amework. This rule maximizes the total reduction in generalization err
or over the whole learning process. A simple example demonstrates that
the locally optimal rule, which maximizes the rate of decrease in gen
eralization error, may perform poorly in comparison.