Static field approach for pattern classification
Authors: Ruta, D. and Gabrys, B.
Journal: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume: 2311
Pages: 232-246
eISSN: 1611-3349
ISBN: 9783540434818
ISSN: 0302-9743
DOI: 10.1007/3-540-46019-5_18
Abstract:Recent findings in pattern recognition show that dramatic improvementfof the recognition rate can be obtained by application of fusion systemsfutilizing many different and diverse classifiers for the same task. Apart from afgood individual performance of individual classifiers the most important factorfis the useful diversity they exhibit. In this work we present an example of afnovel non-parametric classifier design, which shows a substantial level of diversityfwith respect to other commonly used classifiers. In our approach inspirationffor the new classification method has been found in the physical world. Namelyfwe considered training data as particles in the input space and exploited the conceptfof a static field acting upon the samples. Specifically, every single datafpoint used for training was a source of a central field, curving the geometry offthe input space. The classification process is presented as a translocation in thefinput space along the local gradient of the field potential generated by the trainingfdata. The label of a training sample to which it converged during the translocationfdetermines the eventual class label of the new data point. Based on selectedfsimple fields found in nature, we show extensive examples and visual interpretationsfof the presented classification method. The practical applicabilityfof the new model is examined and tested using well-known real and artificialfdatasets.
Source: Scopus
