Asynchronous Parallel Computation of Self-organizing Maps

M.W. Benson (Canada) and J. Hu (USA)

Keywords

neural networks, selforganizing, parallel, asynchronous,optimization

Abstract

An asynchronous algorithm is formulated for generating self organizing maps that arise from the minimization of an energy function. Results using a simulations along with ex periments using a shared memory parallel computer show the asynchronous approach to be effective. Successful re sults are also demonstrated with an asynchronous variation on the Kohonen self-organizing map algorithm where no global energy function is present. A partitioning of the weight array so each processor is responsible for a small part of this array proves effective and improves the parallel efficiency of the asynchronous approach.

Important Links:



Go Back