Η τελευταία διάλεξη του μαθήματος θα γίνει τη Δευτέρα 6 Ιουνίου, ώρα 17:00 - 20:00, στην αίθουσα 1.1.31, στο παλιό κτήριο Ηλεκτρολόγων.
Η διάλεξη θα γίνει από τον Παναγιώτη Μερτικόπουλο (CNRS). Ακολουθούν ο τίτλος και μια σύντομη περίληψη της διάλεξης του κ. Μερτικόπουλου
TITLE. From convex optimization to learning in games
ABSTRACT. The purpose of this talk is to provide a unified view of a broad spectrum of iterative algorithms for problems with a "convex structure" – from good old gradient descent for convex minimization, to algorithms like Hedge and EXP3 for learning in games and multi-armed bandits. The glue that holds this framework together is the so-called "mirror descent" template, a family of algorithmic schemes where ordinary gradient steps and projections are replaced by a "mirror" analogue. We will subsequently use this viewpoint to survey a range of results in the field – both old and new – and if there is time, we will discuss a number of open questions that have attracted considerable interest in machine learning and beyond.