self-organising maps

The trends of the past decade in architecture of the central processing unit show a clear direction towards multi-core processors with the number of cores increasing every eighteen months according to the Moore’s law. The shift from fast single-core to slower multi-core CPUs poses a question of scalability for a vast class of computational algorithms including algorithm used for data analysis. This paper presents the research result of using state of the art parallelisation programming paradigms to scale data analysis processes based on Self-Organising Maps.

In this article we present an algorithm for visualising the clustering structure of the data model captured by dynamic interval self-organising map (DISOM). The developed visualisation algorithm employs the Self-Organising Map for placing DISOM elements on the 2D lattice in conjunction with U-Matrix algorithm for visualization of data clusters.