Biologically Plausible Fuzzy-Knowledge-Out and Its Induced Wide Learning of Interpretable TSK Fuzzy Classifiers

As an alternative to existing construction methods of Takagi-Sugeno-Kang (TSK) fuzzy classifiers, this paper presents a novel design methodology formulated by a new concept called fuzzy-knowledge-out and its induced wide learning way. Analogous to the “dropout” concept in deep learning, the concept of fuzzy-knowledge-out in TSK fuzzy classifiers is motivated by the firing pattern of knowledge in biological neural networks. Our theoretical analysis reveals that a fuzzy classifier built after fuzzy-knowledge-out from a complete set of highly interpretable fuzzy rules is distinctive in generalization and coadaption avoidance. As such, an ensemble called wide learning based TSK (WL-TSK), of highly interpretable zero-order TSK fuzzy subclassifiers constructed quickly by means of fuzzy-knowledge-out operations in a wide learning manner is proposed to achieve enhanced classification performance and high interpretability. With the use of the proposed halving or averaging operations, WL-TSK essentially behaves like only one zero-order TSK fuzzy classifier. Thus, the proposed method can be considered as a new design methodology of TSK fuzzy classifiers. Our experimental results on 15 datasets indicate the effectiveness of WL-TSK in terms of both enhanced classification performance and high interpretability.