posted 2009-10-20T15:54:39Z
bialek et al. and hausser et al. present competing methods for estimating entropy or mutual information. bialek proposes starting with a bayesian prior which is uniform in entropy, whereas hausser talks about a regularization approach which implies cross-validation. when i first read bialek's paper years ago, it was helpful in getting me to start thinking about the difficult estimation issues here, but i like hausser's approach because it seems to work and regularization is so simple.