We present a novel quantization algorithm based on building and adaptively pruning a partition on the data space. The proposed method has connections with decision trees, wavelet representation and can be contrasted to classic quantization schemes such as k-means. The performance of the algorithms are analyzed in a general statistical learning framework where data are assumed to be sample according to un unknown distribution. In particular, the obtained error estimates depend on the geometric properties of the support of the distribution and cover the special case where the latter is a manifold. Joint work with Enrico Cecini (Universita’ di Genova) and Ernesto De Vito (Universita’ di Genova)

This presentation is part of Minisymposium “MS42 - Low dimensional structures in imaging science (3 parts)”
organized by: Wenjing Liao (Georgia Institute of Technology) , Haizhao Yang (Duke University) , Zhizhen Zhao (University of Illinois Urbana-Champaign) .

Authors:

Lorenzo Rosasco (University of Genoa, Istituto Italiano di Tecnologia; Massachusetts Institute of Technology)

Keywords:

machine learning