An estimator of the mutual information based on a criterion for conditional independence

✍ Scribed by Georges A. Darbellay

• Publisher: Elsevier Science
• Year: 1999
• Tongue: English
• Weight: 170 KB
• Volume: 32
• Category: Article
• ISSN: 0167-9473
• DOI: 10.1016/s0167-9473(99)00020-1

✦ Synopsis

The mutual information is a measure of stochastic dependence. Here, we present a data-dependent nonparametric estimator of the mutual information. The algorithm, upon which this estimator rests, is based on Dobrushin's information theorem. The key idea is to build a succession of ÿner and ÿner partitions made of nested hyperrectangles, and stop the reÿnement process on any hyperrectangle as soon as local independence has been achieved. The bias and variance of this estimator are studied through simulations. This includes a comparison to maximum likelihood estimators.