A distributed learning algorithm based on frontier vector quantization and information theory

Abstract

In this paper, we propose a novel distributed learning algorithm built upon the Frontier Vector Quantization based on Information Theory (FVQIT) method. The FVQIT is very effective in classification problems but it shows poor training time performance. Thus, distributed learning is appropriate here to speed up training. One of the most promising lines of research towards learning from distributed data sets is separated learning and model integration. Separated learning avoids moving raw data around the distributed nodes. The integration of local models is implemented in this research using a genetic algorithm. The results obtained from twelve classification data sets demonstrate the efficacy of the proposed method. In average, the distributed FVQIT performs 13.56 times faster than the FVQIT and improves classification accuracy by 5.25%. © 2013 Springer-Verlag Berlin Heidelberg.