Abstract

This work revisits and extends the problem of consistent divergence estimation using data-dependent partitions. For distributions defined on (¿d, B(¿d)), the main result characterizes sufficient conditions on a data-dependent partition scheme to get a strongly consistent histogram-based estimate of the divergence.

Date

June 28, 2009

Authors

Jorge Silva, Shrikanth S Narayanan

Conference

2009 IEEE International Symposium on Information Theory

Pages

468-472

Publisher

IEEE