LANS Informal Seminar: Carlo Graziani

Seminar Title: Better Forecasting Through Information Theory
Speaker: Carlo Graziani, Computational Scientist

Date/Time: 2018-01-10 10:30
Location: Bldg. 240. Rm. 1404-1405

Description:
Probabilistic forecasts are fundamental tools for making decisions under
uncertainty in a wide variety of fields, including weather, energy use,
and finance. I will present a scheme whereby a base probabilistic
forecasting system that is poorly-calibrated may be recalibrated by
incorporating past performance information to produce a new forecasting
system that is demonstrably superior to the original one, in that it can
consistently win wagers against the original system. The scheme utilizes
Gaussian Process modeling to estimate probability measures in a manner
that gives closed-form access to information entropy measures, which
allows prediction of winnings in certain wagers against the base
forecasting system. The recalibration scheme is formulated in a
framework that exploits the deep connections between information theory,
forecasting, and betting.