Efficient Implementation of the THSOM Neural Network

Authors:
Rudolf Marek;Miroslav Skrbek
Affiliations:
Department of Computer Science and Engineering, Faculty of Electrical Engineering, Czech Technical University in Prague, Prague, Czech Republic 121 35;Department of Computer Science and Engineering, Faculty of Electrical Engineering, Czech Technical University in Prague, Prague, Czech Republic 121 35
Venue:
ICANN '08 Proceedings of the 18th international conference on Artificial Neural Networks, Part II
Year:
2008

Citing 3
Cited 0

Self-Organizing Maps

Self-Organizing Maps
parSOM: A Parallel Implementation of the Self-Organizing Map Exploiting Cache Effects: Making the SOM Fit for Interactive High-Performance Data Analysis

IJCNN '00 Proceedings of the IEEE-INNS-ENNS International Joint Conference on Neural Networks (IJCNN'00)-Volume 6 - Volume 6
Hardware Accelerated Data Analysis

PARELEC '04 Proceedings of the international conference on Parallel Computing in Electrical Engineering

Quantified Score

Hi-index	0.00

Visualization

Abstract

Recent trends in microprocessor design clearly show that the multicore processors are the answer to the question how to scale up the processing power of today's computers. In this article we present our C implementation of the Temporal Hebbian Self-organizing Map (THSOM) neural network. This kind of neural networks have growing computational complexity for larger networks, therefore we present different approaches to the parallel processing --- instruction based parallelism and data-based parallelism or their combination. Our C implementation of THSOM is modular and multi-platform, allowing us to move critical parts of the algorithm to other cores, platforms or use different levels of the instruction parallelism yet still run exactly the same computational flows --- maintaining good comparability between different setups. For our experiments, we have chosen a multicore x86 system.