An adaption technique to SVD reduced rule bases

The practically non-universal approximation property, shown by Tikk et al. and the exponential complexity problem of widely adopted fuzzy logic techniques, shown by Koczy and Hirota, reveal the contradictions features of fuzzy rule bases in pursuing of good approximation. As a result complexity reduction topic emerged in fuzzy theory. One of the natural disadvantages of using complexity reduction is that the adaptivity property of the reduced approximation becomes strictly restricted. This paper proposes a technique, to singular value decomposition (SVD) based reduction, which may alleviate the adaptivity restriction. A high order tensor projection is proposed here as key idea.