Constrained Clustering: Effective Constraint Propagation with Imperfect Oracles

While clustering is usually an unsupervised operation, there are circumstances in which we have prior belief that pairs of samples should (or should not) be assigned to the same cluster. Constrained clustering aims to exploit this prior belief as constraint (or weak supervision) to influence the cluster formation so as to obtain a structure more closely resembles human perception. Two challenging issues remain open: (1) how to propagate sparse constraint effectively, and (2) how to handle ill-conditioned/noisy constraint generated by imperfect oracles. In this paper we present a unified framework to address the above issues. Specifically, in contrast to existing constrained clustering approaches that blindly rely on all features for propagation, our approach searches for neighbours driven by discriminative feature selection for more effective constraint diffusion. Crucially, we formulate a novel data-driven filtering approach to handle the noisy constraint problem, which has been unrea...

• Type: Conference
• Year: 2013
• Where: ICDM
• Authors: X. Zhu, C. C. Loy, and S. Gong
• Keywords: Clustering, Constrained Clustering, Manifold, Noisy Oracle, Spectral Clustering