Conference article

Adaptive Robust SVM-Based Classification Algorithms for Multi-Robot Systems using Sets of Weights

Lev V. Utkin
Telematics Department, Peter the Great St.Petersburg Polytechnic University, Russia

Vladimir S. Zaborovsky
Telematics Department, Peter the Great St.Petersburg Polytechnic University, Russia

Sergey G. Popov
Telematics Department, Peter the Great St.Petersburg Polytechnic University, Russia

Download articlehttp://dx.doi.org/10.3384/ecp17142959

Published in: Proceedings of The 9th EUROSIM Congress on Modelling and Simulation, EUROSIM 2016, The 57th SIMS Conference on Simulation and Modelling SIMS 2016

Linköping Electronic Conference Proceedings 142:141, p. 959-965

Show more +

Published: 2018-12-19

ISBN: 978-91-7685-399-3

ISSN: 1650-3686 (print), 1650-3740 (online)

Abstract

Three adaptive iterative minimax multi-robot system learning algorithms are proposed under condition that every observation obtained from robots is set-valued, i.e., it consists of several elements. The set-valued data are caused due to a fact that robots in the system provide different measurements in a single system observation. The ?rst idea underlying the algorithms is to use sets of weights or imprecise weights of a special form for all elements of training data. The second idea is to apply the imprecise Dirichlet model for iterative updating the sets of weights depending on the classi?cation accuracy and for assigning new weights to robots for improving classi?ers. The simplest ?rst algorithm is a modi?cation of the SVM in order to take into account set-valued data. The second algorithm is the AdaBoost with the modi?ed SVM under set-valued data. The third algorithm is the modi?cation of the AdaBoost with updating imprecise weights of robots. The algorithms allow us to take into account the set-valued observations in the framework of the minimax decision strategy and to get optimal weights of robots to improve the classi?cation accuracy of the trained multi-robot system.

Keywords

multi-robot system, SVM, classi?cation, AdaBoost, set-valued observations, sets of weights

References

No references available

Citations in Crossref