Speaker: Lior Horesh, IBM TJ Watson
Location: Warren Weaver Hall 1302
Date: April 21, 2017, 10 a.m.
Mathematical models are employed ubiquitously for description, prediction and decision making. In addressing end-goal objectives, great care needs to be devoted to attainment of appropriate balance of inexactness throughout the various stages of the underlying process. While “all models are wrong...”, it is often possible to deduce a correction for a model, that offers improved predictive and / or prescriptive capabilities. In this talk we shall present a hybrid first-principles, data-driven stochastic optimization framework for model mis-specification correction. The approach balances between model fidelity, model complexity, and virtues associated with both the correction and the corrected forms.