A novel Continuous Action-set Learning Automaton algorithm

作者:Guo, Ying*; Ge, Hao; Yan, Yan; Huang, Yuyang; Li, Shenghong
来源:1st International Conference on Signal and Information Processing, Networking and Computers (ICSINC), China,Beijing, 2015-10-17 to 2015-10-18.

摘要

A novel Continuous Action-set Learning Automaton algorithm named Adaptive Sampling CALA (ASCALA) is presented in this paper. In the proposed model, learning process is a combination of the sampling phase and the iteration phase. The new philosophy lies in the acquisition of a priori knowledge after sampling sufficiently. The learning step and the variance of action selection probability are adaptively adjusted during the two phases, varying from a small-step to a large-step learning and a large invariance to a gradually smaller one. The experiments with regard to function optimization evidently showed that our proposed method outshone the original algorithm regardless of the initial value of parameters. Besides, our algorithm is robust to the noise that added to the function.