Optimizing the parSOM neural network implementation for data mining with distributed memory systems and cluster computing

Abstract

The self-organizing map is a prominent unsupervised neural network model which lends itself to the analysis of high-dimensional input data and data mining applications. However; the high execution times required to train the map put a limit to its application in many high-performance data analysis application domains.In this paper we discuss the ,,&OM imp!ementation, a software-based parallel implementation of the seljorganizing map, and its optimization for the analysis of high-dimensional input data using distributed memory systems and clusters. The original r,&OM algorithm scales very well in a parallel execution environment with low communication latencies and exploits parallelism to cope with memory latencies. However it suffers from poor scalability on distributed memory computers. We present optimizations to further decouple the subprocesses, simplifi the conimunication model and improve the portabilit?, of the sj'stem. 0-7695-0680-1/00 $10.00 0 2000 IEEE

Mining massive datasets by an unsupervised parallel clustering on a GRID: Novel algorithms and case study

“…(a) the network partitioning schemes are more suitable for a multiprocessor environment where the communication overhead may be greatly reduced thanks to the high speed of the common bus shared by the processors, as shown in [16,30,31]; (b) the data partitioning schemes become appropriate when the parallel SOM is executed in loosely coupled systems, as shown in [32] and in [33], although the proposed implementations exploit only 60%-80% of the overall speed-up due to the increase of the communication overhead at the increase of the number of CEs [34].…”

“…P2 = 0. 30 The results for the dataset which will be used also in the case study are reported in Table 9 where the items were clustered following the F1 similarity by the WR WM PSOM executed by 9 slaves. It is important to note that the positive feature matching in average between the sequential and the parallel SOM is very high (about 90%-94%).…”

Application of Self Organizing Map (SOM) to model a machining process

“…These wide varieties of examples demonstrate that considerable effort has been placed on applying these multivariate tools. Optimizations to par SOM, a software-based parallel implementation of the SOM was first introduced by Tomsich et al (2000) which provides a better performance compared to other implementation attempts such as the one reported in Boniface et al (1999).…”

Design and Implementation of Parallel Counterpropagation Networks Using MPI

“…A parallel neural network can be constructed using a variety of different methods (Standish, 1999;Schikuta, 1997;Serbedzija, 1996;Schikuta et al, 2000;Misra, 1992;Misra, 1997), such as the parallel virtual machines (PVM) (Quoy, 2000), the message passing interface (MPI) (Snir et al, 1998;Cropp et al, 1998;Pacheco, 1997), the shared memory model and the implicit parallelization with parallel compiler directives . Concerning the network types that have been paralellized by one of these methods, they cover a very broad range from the supervised back propagation network (Torresen et al, 1994;Torresen and Tomita, 1998;Kumar, 1994) to the unsupervised self-organizing maps (Weigang et al, 1999;Tomsich et al, 2000). In this research the counterpropagation network is parallelized by means of the message passing interface library (Pacheco, 1997).…”