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

This paper addresses the problem of pattern classification using neural networks. Applying neural network classifiers for classifying a large volume of high dimensional data is a difficult task as the training process is computationally expensive. A parallel implementation of the known training paradigms offers a feasible solution to the problem. By exploiting the massively parallel structure of the LevenbergMarquardt 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 (Message Passing Interface). The implementation, besides facilitating the main objective of maximising computational speedup, is also portable and scalable. A standard benchmark for neural network training comprising a sufficiently large volume of satellite image data has been utilized to present and discuss the properties of the implementation.