Automatically weighted binary multi-view clustering via deep initialization (AW-BMVC)

Clustering is inherently a process of exploratory data analysis. It has attracted more attention recently because much real-world consists multiple representations or views. However, it becomes increasingly problematic when dealing with large and heterogeneous data. worth noting that several approaches have been developed to increase computational efficiency, although most them some drawbacks: 1) Most existing techniques consider equal static weights quantify importance across different views samples, so common complementary features cannot be used. 2) The clustering task performed by arbitrary initialization without caring about the rich structure joint discrete representation, thus poorly executed. In this paper, we propose novel approach called ”Auto-Weighted Binary Multi-View Via Deep Initialization” for large-scale multi-view based on two main scenarios. First, distinction between therefore apply dynamic learning strategy automatic weighting samples. Second, in context initializing binary clustering, develop new CNN feature use low-dimensional embedding exploiting efficient capabilities Fourier mapping. Moreover, our simultaneously learns representation performs direct using constrained matrix factorization; optimization problem perfectly solved unified model. Experimental results conducted challenging datasets demonstrate effectiveness superiority proposed over state-of-the-art methods terms accuracy, normalized mutual information, purity.