The Optimization of Kernel CMAC Based on BYY Learning

Cerebellar Model Articulation Controller (CMAC) has attractive properties of fast learning and simple computation. The kernel CMAC, which provides an interpretation for the classic CMAC from the kernel viewpoint, strengthens the modeling capability without increasing its complexity. However, the kernel CMAC suffers from the problem of selecting its hyperparameter. In this paper, the Bayesian Ying-Yang (BYY) learning theory is incorporated into kernel CMAC, referred to as KCMAC-BYY, to optimize the hyperparameter. The BYY learning is motivated from the well-known Chinese Taoism Yin-Yang philosophy, and has been developed in this past decade as a unified statistical framework for parameter learning, regularization, structural scale selection and architecture design. The proposed KCMAC-BYY achieves the systematic tuning of the hyperparameter, further improving the performance in modeling capability and stability. The experimental results show that the proposed KCMAC-BYY outperforms the existing representative techniques in the research literature.