Artificial Neural Networks in Classifying Telemetry Data for Compression Systems
Abstract
Introduction: Data compression systems with algorithmic adaptation always contain a certain classifier which chooses the most efficient way to compress the input data. Traditionally, the choice of estimation characteristics for the classifier is based on expert opinion, and this can lead to a worse quality of classification when processing data with complex structure. Furthermore, for such systems you have to provide parallelism of the procedures which train the qualifier and compress the data. This increases the computational cost and complicates the architecture of the transceivers. Thus, the problem of developing an efficient qualifier for compression systems is a pressing issue. Purpose: The aim of the research is to estimate the possibility of using a neural multilayer direct-propagation network with a given architecture as a telemetry data classifier Results: We studied the behavior of averaged errors in training, generalization and acknowledgement depending on the size of the learning sample obtained for several sets of telemetry data. On the base of the obtained data, optimal parameters of neural network training have been offered. The proposed approach has been compared in terms of efficiency to some specialized numerical methods, namely: background, entropy, neighborhood and dihedral methods. The results lead to the conclusion that even the simplest and most versatile architecture of an artificial neural network significantly exceeds the systems with numerical methods of estimating the classifying attribute. Further development of this approach involves a more indepth analysis of neural network architecture in order to generate optimal configurations for various amounts of training samples and develop specialized neural network architectures for specified data types.Published
2016-06-01
How to Cite
Bogachev, I., Levenets, A., & Chye, E. (2016). Artificial Neural Networks in Classifying Telemetry Data for Compression Systems. Information and Control Systems, (3), 2-7. https://doi.org/10.15217/issn1684-8853.2016.3.2
Issue
Section
Information processing and control