“…MVL seeks to employ multiple "independent" clues (e.g., bi-lingual information, different modalities); MKL, however, combines multiple base kernels to create a "unified" kernel for learning, where these kernels are not necessarily independent like the "views" in MVL. Similarly, both were originally studied for (semi-)supervised learning, and have been extended to unsupervised setting recently [33,6,14,15,17,16]. Multiview clustering [20,34,15,16], as an extension of MVL, assumes inherently that the views are uncorrelated.…”