Distributed Implementation of NNSU Separating Algorithm

0366087 - ÚI 2012 RIV CZ eng L4 - Software
Hakl, František
Distributed Implementation of NNSU Separating Algorithm.
Internal code: WWW-NNSU ; 2011
Technical parameters: Paralelni implementace separačního algoritmu geneticky optimalizovaných neuronových sítí s přepínacími jednotkami. Webové rozhraní této implementace pro vzdálené využívání na bázi klient-server.
Economic parameters: umožňuje separaci datových množin na základě známých (měřených) vzorů
R&D Projects: GA TA ČR TA01010490; GA MŠMT(CZ) 1M0567
Institutional research plan: CEZ:AV0Z10300504
Keywords : separation * prediction * data mining * neural networks * genetic optimization
Subject RIV: IN - Informatics, Computer Science

Neural Network with Switching Units (NNSU) is a feed-forward neural network model. Nodes of this network are conected to form general Directed Acyclic Graph (DAG). The structure of this DAG is set up by genetic optimization procedures based on cross-over consistent representation of DAG structure. Nodes of the network consist from computational units and switching units. Switching units split an input space into separated parts, depending on results of previous computational units. Data in such subsets of input space are localy separated by consequent computational units. This model provides better separation/approximation ability than clasical neural networks approach. In addition, the fitness function of the genetic algorithm can by tuned to meet user specific requirements.
Permanent Link: http://hdl.handle.net/11104/0201178