Host: Japan SOciety for Fuzzy Theory and intelligent informatics
Co-host: The Korea Fuzzy Logic and Intelligent Systems Society, IEEE Computational Intelligence Society, The International Fuzzy Systems Association, 21th Century COE Program "Creation of Agent-Based Social Systems Sciences"
This paper presents construction of an adaptive neural network ensemble i.e., try to solve the given problem with single neural network and build ensemble with minimum architecture when required. Exiting methods always think multiple networks for ensemble case though some problem may solvable by single network. This new method utilizes constructive approach to determine ensemble size automatically where it adds networks one by one and uses cumulative number of hidden nodes for coming networks. Also, at the time of network addition, a new network is motivated on previously unsolved portion of training space. Finally all the networks are trained simultaneously. This new method has been tested extensively on several benchmark problems of machine learning and neural networks. Experi-mental results show that this method able to construct adaptive ensemble in which some problem is solved by single neural network and for multiple networks case used minimum ensemble structure.