摘要: In this paper we continue study of the games prediction with expert advice uncountably many experts. A convenient interpretation such is to construe pool experts as one “stochastic predictor”, who chooses in at random according prior distribution on and then replicates (deterministic ) predictions chosen expert. We notice that if stochastic predictor‘s total loss most L probability least p learner‘s can be bounded by cL + aln \frac{1}{p} for usual constants c a. This used revamp known results obtain new tracking best It also applied merging overconfident fitting polynomials data.