Title :
A motivation-following learning
Author :
Zhou, SU ; Popovic, Dobrivoe ; Schulz-Ekloff, Guenter
Author_Institution :
Inst. of Autom. Technol., Bremen Univ., Germany
Abstract :
Concerns a learning process in which the objective is to follow a primary motivation and is usually represented by the minimization of a sum-of-the-squared error. A motivation-following learning, especially suitable for the learning of a rapidly changing environment, is proposed to update the parameter vector of the modified multiblock network. Its advantages are online adaptation ability, simplicity of implementation, and steadily improved learning performance.
Keywords :
backpropagation; least squares approximations; minimisation; neural nets; implementation simplicity; modified multiblock network; motivation-following learning; online adaptation ability; rapidly changing environment; sum-of-the-squared error minimization; Acceleration; Automation; Backpropagation algorithms; Bismuth; Chemical technology; Chemistry; Convergence; Jacobian matrices; Learning systems; Niobium;
Conference_Titel :
Neural Networks, 1993. IJCNN '93-Nagoya. Proceedings of 1993 International Joint Conference on
Print_ISBN :
0-7803-1421-2
DOI :
10.1109/IJCNN.1993.713974
