Teaching Feed-Forward Neural Networks by Simulated Annealing
Jonathan Engel
Norman Bridge Laboratory of Physics 161-33, California Institute of Technology,
Pasadena, CA 91125, USA
Abstract
Simulated annealing is applied to the problem of teaching feed-forward neural networks with discrete-valued weights. Network performance is optimized by repeated presentation of training data at lower and lower temperatures. Several examples, including the parity and "clump-recognition" problems are treated, scaling with network complexity is dicussed, and the viability of mean-field approximations to the annealing process is considered.
