Implementations of asynchronous self-organizing maps on OpenMP and MPI parallel computers

In [1], we presented an asynchronous parallel algorithm for self-organizing maps based on a recently defined energy function which leads to a self-organizing map. We generalized the existing stochastic gradient approach to an asynchronous parallel stochastic gradient method for generating a topological map on a distributed computer system (MIMD). We theoretically proved that our algorithm was convergent and the simulations showed our algorithm was effective. In this paper, we implement this algorithm on practical parallel computers with two different types: openMP and MPI, in the Supercomputing Institution at University of Minnesota. By analyzing the experimental results, we demonstrate the convergence, efficiency and speed-up of our algorithm.