Sparsity-Inducing Fuzzy Subspace Clustering
Arthur Guillon, Marie‐Jeanne Lesot, Christophe Marsala · Repository KITopen (Karlsruhe Institute of Technology) · 2018
This paper considers a fuzzy subspace clustering problem and proposes to introduce an original sparsity-inducing regularization term. The minimization of this term, which involves a l$_{0}$ penalty, is considered from a geometric point of view and a novel proximal operator is derived. A subspace clustering algorithm, Prosecco, is proposed to optimize the cost function using both proximal and alternate gradient descent. Experiments comparing this algorithm to the state of the art in sparse fuzzy subspace clustering show the relevance of the proposed approach.