| Summary: | 碩士 === 國立臺灣科技大學 === 電機工程系 === 90 === A Robust Fuzzy Credit Assignment CMAC (FCA-CMAC) is proposed in this thesis to speed up the learning process and to increase the robust capability of CMAC. The Fuzzy Credit assigned CMAC is to combine the fuzzy logic concept and credit assignment ideas to provide fast and accurate learning for CMAC. With fuzzified blocks, CMAC can increase its precision and resolution; Meanwhile, the calculated errors are assigned proportional to the inverse of learning times, which are viewed as creditabilities of the addressed hypercubes. Besides, we also embed the robust learning algorithm (i.e. M-estimators) concept into the CMAC learning algorithm. The basic idea of M-estimators is to use a loss function to degrade the effects of outliers. In the simulation process, we also propose to tune the learning constant as α=(1/epoch) and to select the scalar estimator β=0.1 so as to have better learning performances. From simulations, it can be seen that in the off-line learning, FCA-CMAC has the fastest learning capability and in the on-line learning, the proposed FCA-CMAC can also get best performances.
|
|---|
