A parallel variable memory BFGS training algorithm

This paper considers the parallel implementation of a novel variable memory quasi-newton neural network training algorithm recently developed by the author. Unlike existing training methods this new technique is able to optimize performance in relation to available memory. Numerically it has equivalent properties to Full Memory BFGS optimization (FM) when there are no restrictions on memory and to FM with periodic reset when memory is limited. Parallel implementations of both the Full and Variable Memory BFGS algorithms are outlined and performance results presented for a PVM target architecture. Copyright (C) 2000 IFAC.