Abstract
Unsupervised multi-view representation learning has been extensively studiedfor mining multi-view data. However, some critical challenges remain. On theone hand, the existing methods cannot explore multi-view data comprehensivelysince they usually learn a common representation between views, given thatmulti-view data contains both the common information between views and thespecific information within each view. On the other hand, to mine the nonlinearrelationship between data, kernel or neural network methods are commonly usedfor multi-view representation learning. However, these methods are lacking ininterpretability. To this end, this paper proposes a new multi-view fuzzyrepresentation learning method based on the interpretable Takagi-Sugeno-Kang(TSK) fuzzy system (MVRL_FS). The method realizes multi-view representationlearning from two aspects. First, multi-view data are transformed into ahigh-dimensional fuzzy feature space, while the common information betweenviews and specific information of each view are explored simultaneously.Second, a new regularization method based on L_(2,1)-norm regression isproposed to mine the consistency information between views, while the geometricstructure of the data is preserved through the Laplacian graph. Finally,extensive experiments on many benchmark multi-view datasets are conducted tovalidate the superiority of the proposed method.