Parallel Levenberg-Marquardt-Based Neural Network Training on Linux Clusters-A Case Study.

Abstract

This paper addresses the problem of pattern classification
\nusing neural networks. Applying neural network classifiers
\nfor classifying a large volume of high dimensional data is
\na difficult task as the training process is computationally
\nexpensive. A parallel implementation of the known train-
\ning paradigms offers a feasible solution to the problem. By
\nexploiting the massively parallel structure of the Levenberg-Marquardt algorithm for non-linear optimization a training algorithm for neural networks has been implemented on a Linux cluster using LAM (Local Area Multi-computer) MPI
\n(Message Passing Interface). The implementation, besides
\nfacilitating the main objective of maximising computational
\nspeedup, is also portable and scalable. A standard bench-
\nmark for neural network training comprising a sufficiently
\nlarge volume of satellite image data has been utilized to
\npresent and discuss the properties of the implementation