Title :
Aspects of learning neural control
Author :
Riedmiller, Martin
Author_Institution :
Inst. fur Logik, Karlsruhe Univ., Germany
Abstract :
The possibilities and restrictions for applying neural architectures to control problems are discussed in the light of available a priori knowledge. Reinforcement learning is shown to provide a useful solution in the absence of any a priori knowledge. Two reinforcement learning architectures are proposed and the quality of the incrementally acquired control strategy is discussed on the well known cart pole benchmark problem
Keywords :
learning (artificial intelligence); neurocontrollers; cart pole benchmark problem; incrementally acquired control strategy; learning neural control; neural architectures; reinforcement learning; Computer architecture; Control systems; Control theory; Fuzzy control; Fuzzy systems; Humans; Knowledge based systems; Learning; Lighting control; Neural networks; Supervised learning;
Conference_Titel :
Systems, Man, and Cybernetics, 1994. Humans, Information and Technology., 1994 IEEE International Conference on
Conference_Location :
San Antonio, TX
Print_ISBN :
0-7803-2129-4
DOI :
10.1109/ICSMC.1994.400051
