Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: Modal-set estimation with an application to clustering
(Submitted on 13 Jun 2016)
Abstract: We present a first procedure that can estimate -- with statistical consistency guarantees -- any local-maxima of a density, under benign distributional conditions. The procedure estimates all such local maxima, or $\textit{modal-sets}$, of any bounded shape or dimension, including usual point-modes. In practice, modal-sets can arise as dense low-dimensional structures in noisy data, and more generally serve to better model the rich variety of locally-high-density structures in data.
The procedure is then shown to be competitive on clustering applications, and moreover is quite stable to a wide range of settings of its tuning parameter.
Link back to: arXiv, form interface, contact.