Spaks: Self-paced multiple kernel subspace clustering with feature smoothing regularization

Highlights：

• We propose a new multiple kernel subspace clustering method-SPAKS to address the high-dimensional nonlinear data clustering problem.

• We propose a feature smoothing regularization for the kernel subspace clustering model to facilitate the smoothness of the feature representation after kernel mapping on the affinity graph, thus ensuring that the data points between different subspaces are sufficiently separable.

• We adopt the self-paced learning framework to optimize the weights of the base kernels in the multiple kernel scenario, which can fully exploit the fitting ability of each base kernel and make the model solving process easier.

• We conduct extensive experiments on seven benchmark data sets to validate the superiority of the proposed SPAKS method on the clustering task.