Density estimation consistent in information divergence
László Györfi, Edward C. van der Meulen · 2002
We consider the problem of estimating an unknown density function f in information divergence. We prove the following theorem: for any sequence of density estimates {f/sub n/} there is a density f with finite differential entropy H(f) and with arbitrarily many derivatives such that D(f, f/sub n/)=/spl infin/ for all n.>