О подходе к выявлению многократной разладки в испытаниях Бернулли с использованием CTW-предиктора

Abstract

"An approach to statistical test construction for randomness testing of binary sequences on the base of universal predictors is considered. We derive the power of statistical tests constructed on the base of the maximum-likelihood predictors for the models of Bernoulli asymmetric trials and Markov chain. The results are extended to the Context Tree Weighting predictor, which is universal for these models. We show that usage of universal predictor in sliding window allows to detect effectively multiple model changes. Comparison of the proposed approach and traditional methods of multiple changes detection is conducted."