Welcome!
To use the personalized features of this site, please log in or register.
If you have forgotten your username or password, we can help.
|
 |
parSOM: Using Parallelism to Overcome Memory Latency in Self-Organizing Neural Networks
| |
|
parSOM: Using Parallelism to Overcome Memory Latency in Self-Organizing Neural Networks
Philipp Tomsich7  , Andreas Rauber7 and Dieter Merkl7
| (7) |
Institute of Software Technology, Vienna University of Technology, Favoritenstraße 9-11/188, A-1040 Wien, Austria |
Abstract
The self-organizing map is a prominent unsupervised neural network model which lends itself to the analysis of high-dimensional
input data. However, the high execution times required to train the map put a limit to its application in many high-performance
data analysis application domains, where either very large datasets are encountered and/or interactive response times are
required.
In this paper we present the parSOM, a software-based parallel implementation of the self-organizing map, which is particularly optimized for the analysis
of high-dimensional input data. This model scales well in a parallel execution environment, and, by coping with memory latencies,
a better than linear speed-up can be achieved using a simple, asymmetric model of parallelization. We demonstrate the benefits
of the proposed implementation in the field of text classification, which due to the high dimensionalities of the data spaces
encountered, forms a prominent application domain for high-performance computing.
Fulltext Preview (Small, Large)
 References secured to subscribers.
|
|
|
|
|
|