Parallel Implementations of Self-Organizing Maps
Timo D. Hämäläinen · Studies in fuzziness and soft computing · 2002
This chapter focuses on parallel implementations of the Self-Organizing Map (SOM) featuring different levels of parallelism. The basic arithmetic-logical operations of SOM are first reviewed for a consideration of implementation issues such as number precision, memory consumption and time complexity. Mapping involves network , training set , neuron and weight parallelism. Examples of the weight and neuron parallel mappings are given for abstract platforms to conduct general principles. Neuron parallel mapping is considered in great detail as it is the most commonly used approach. A review of implementations is given from supercomputers to VLSI (Very Large Scale Integration) chips with criteria for performance comparison. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.