Abstract

A modified quasi-Newton algorithm for training large-scale feedforward neural networks is developed in this paper. It is a vector based training algorithm derived from full memory BFGS and has only O(n) memory usage. Compared with quasi-Newton method, the proposed algorithm requires less memory but convergences with the same order of the speed. The proposed methodology is applied to modeling of industrial product quality with 32-input variables. Simulation results show the effectiveness of the approach.

Paper Details