2 March 2017, Computational Social Choice Seminar, Michail Mamakos

Abstract

In this talk, I will present my work on combining cooperative game theory with graphical models, in order to tackle agent uncertainty regarding the underlying collaboration structure (that naturally gives rise to coalitional values). Moreover, I will show how this work can be applied in settings where coalitions overlap (unlike what is commonly assumed in the literature). In more detail, I will begin by introducing Relational Rules, a concise representation that extends the well-known MC-net representation to cooperative games with overlapping coalitions. I will then proceed to show how an agent can learn topics that correspond to profitable coalitions, by interpreting formed coalitions as documents, and employing online Latent Dirichlet Allocation, a popular probabilistic topic model. This is joint work with my advisor, Georgios Chalkiadakis (Technical University of Crete, Chania).