סמינר מחלקתי

Online Stochastic Optimization with a Moving Target 

Prof. Assaf Zeevi

 Abstract :

Stochastic approximation refers to a family of methods whose objective is to  sequentially estimate the optimum point of a fixed yet unknown cost function whose observations are confounded by statistical noise. Since its inception in the 1950's, the method has been the subject of a voluminous literature and is widely used in a variety of application settings.  In this talk we will  develop some theory that attempts to expand the scope of stochastic approximation to non-stationary environments, i.e., when the cost function is allowed to change over time. Some (hopefully interesting) connections will be drawn to a recent strand of literature on online convex optimization, which studies the aforementioned  problem in a so-called adversarial setting.

Joint work with Omar Besbes, Columbia University, and Yoni Gur, Stanford University. 

ההרצאה תתקיים ביום שלישי 20.1.15, בשעה 14:00 בחדר 206, בנין וולפסון הנדסה, הפקולטה להנדסה, אוניברסיטת תל-אביב.