Universality of Bayesian mixture predictors

MaRDI QIDQ4645636zbMATH OpenFDO

Authors Daniil Ryabko

Publication date 10 January 2019

Full work available at URL https://arxiv.org/abs/1610.08249, http://proceedings.mlr.press/v76/ryabko17a.html

Mathematics Subject Classification ID

Prediction theory (aspects of stochastic processes) (60G25) Bayesian problems; characterization of Bayes procedures (62C10) Minimax procedures in statistical decision theory (62C20) Sequential statistical analysis (62L10)

Abstract: The problem is that of sequential probability forecasting for finite-valued time series. The data is generated by an unknown probability distribution over the space of all one-way infinite sequences. It is known that this measure belongs to a given set C, but the latter is completely arbitrary (uncountably infinite, without any structure given). The performance is measured with asymptotic average log loss. In this work it is shown that the minimax asymptotic performance is always attainable, and it is attained by a convex combination of a countably many measures from the set C (a Bayesian mixture). This was previously only known for the case when the best achievable asymptotic error is 0. This also contrasts previous results that show that in the non-realizable case all Bayesian mixtures may be suboptimal, while there is a predictor that achieves the optimal performance.

Recommendations

• On asymptotic and finite-time optimality of Bayesian predictors
• Convergence and loss bounds for bayesian sequence prediction
• 10.1162/1532443041827952
• Universal time-series forecasting with mixture predictors
• Theory and Applications of Models of Computation

Cited in

(14)

• Universal Bayes consistency in metric spaces
• On universal prediction and Bayesian confirmation
• Asymptotically minimax prediction in infinite sequence models
• Universality of Bayesian predictions
• Rejoinder: ``Universality of Bayesian predictions
• Things Bayes can't do
• On finding predictors for arbitrary families of processes
• On the Consistency of Discrete Bayesian Learning
• Computable Bayesian compression for uniformly discretizable statistical models
• scientific article; zbMATH DE number 1344902 (Why is no real title available?)
• Optimal sequential probability assignment for individual sequences
• 10.1162/1532443041827952
• Universal time-series forecasting with mixture predictors
• On asymptotic and finite-time optimality of Bayesian predictors

This page was built for publication: Universality of Bayesian mixture predictors

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4645636)