Title :
Kalman filter based estimation of dynamic modular networks
Author :
Kadirkamanathan, V. ; Kadirkamanathan, M.
Author_Institution :
Dept. Autom. Control & Syst. Eng., Sheffield Univ., UK
Abstract :
Kalman filter based recursive estimation algorithms for dynamic modular RBF networks have been developed. Two types of algorithms are developed in particular, the first using time information and the second using state information, for modelling systems with multi-modal behaviour. For both of these, “hard” and “soft” competition based estimation schemes are outlined. The estimation algorithms are applied to the problem of learning inverse kinematics and their successful operation is demonstrated
Keywords :
Bayes methods; Kalman filters; feedforward neural nets; learning (artificial intelligence); real-time systems; recursive estimation; Kalman filter; dynamic modular networks; inverse kinematics; learning; modelling; parameter estimation; radial basis function networks; recursive estimation; state information; time information; Automatic control; Heuristic algorithms; Hidden Markov models; Iterative algorithms; Kinematics; Least squares approximation; Neural networks; Nonlinear dynamical systems; Radial basis function networks; Recursive estimation;
Conference_Titel :
Neural Networks for Signal Processing [1996] VI. Proceedings of the 1996 IEEE Signal Processing Society Workshop
Conference_Location :
Kyoto
Print_ISBN :
0-7803-3550-3
DOI :
10.1109/NNSP.1996.548348
