Transfer Representation Learning With TSK Fuzzy System

Transfer learning can address the tasks of unlabeled data in target domain by leveraging plenty labeled from a different but related source domain. A core issue transfer is to learn shared feature space where distributions two domains are matched. This process be named as representation (TRL). Feature transformation methods crucial ensure success TRL. The most commonly used method TRL kernel-based nonlinear mapping high-dimensional space, followed linear dimensionality reduction. But kernel functions lack interpretability, and it difficult select functions. To this end, article proposes more intuitive interpretable method, called with TSK-FS (TRL-TSK-FS), combining TSK fuzzy system (TSK-FS) learning. Specifically, TRL-TSK-FS realizes aspects. On one hand, transformed into distribution distance between minimized. other discriminant information geometric properties preserved analysis principal component analysis. further advantage that realized proposed constructing antecedent part instead functions, which selected. Extensive experiments conducted on text image datasets demonstrate superiority method.