Fuyuhiko TANAKA

Associate Professor
Graduate School of Information Science and Technology, The University of Tokyo

Research Theme

Universal Properties of Statistical Model Manifolds and Its Applications to Bayesian Prediction Theory

Suppose that sequential data is generated according to a statistical model specified by the unknown parameter. Conventional way to estimate the future behavior of the data generating process is to estimate the unknown parameter from the data and at most to predict the forthcoming value and to obtain its confidence interval. In contrast, we consider statistical inference in the framework of statistical prediction, focus on the geometrical properties of statistical models, and aim at construction of the universal theory of giving a better prediction method. Our subject covers very large field, i.e., from time series analysis to qunatum information.