The objective of this paper is to find an estimator for differential entropy. We restrict ourselves to densities that have a bounded support set, that are bounded, and continuous almost everywhere. The techniques on which the proposed estimator is based are very similar to those that play a role in the context-tree weighting method [1995], the difference is that instead of a context tree we use a decomposition tree here. Another difference is that the weighting method includes uncoded coding probabilities. An interesting property of our decomposition-tree weighting estimator is its small computational complexity. We show that our estimator converges to differential entropy with probability one.
展开▼
