Multiple Kernel <i>k</i>-Means With Low-Rank Neighborhood Kernel

Multiple kernel clustering algorithms achieve promising performances by exploring the complementary information from matrices corresponding to each data view. Most of existing methods aim construct a consensus for afterward clustering. However, they ignore that desired is supposed reveal cluster structure among samples and thus be low rank. As consequence, performance could decrease. To address this issue, we propose low-rank learning approach multiple Specifically, instead regularizing with constraints, use re-parameterize scheme matrix. Meanwhile, located in neighborhood area linear combination base kernels. An alternate optimization strategy designed solve resulting problem. We evaluate proposed method on 13 benchmark datasets 9 state-of-the-art algorithms. demonstrated experimental results, our algorithm achieves superior scores against compared reported popular datasets.