Learning in neural networks with local minima
Tom Heskes, Eddy T. P. Slijpen, Bert Kappen · Physical Review A · 1992
An attempt is made to study learning in neural networks with local minima. For small learning parameters \ensuremath{\eta}, the transition time from one mimimum to another is asymptotically given by exp(\ensuremath{\eta}\ifmmode \tilde{}\else \~{}\fi{}/\ensuremath{\eta}), with \ensuremath{\eta}\ifmmode \tilde{}\else \~{}\fi{}, a constant independent of \ensuremath{\eta}, called the reference learning parameter. A general scheme to calculate the reference learning parameter is presented. This scheme is valid for a large class of learning rules.