Speaker: Vessela Daskalova
Title: Learning Categorizations of Strategy Spaces
Authors: Vessela Daskalova, Nicolaas J. Vriend
Abstract: We consider categorization of the strategy space in games, in which some attributes of the game can serve as a basis for distinguishing the available strategies. There may be many different ways to partition the strategy space in a given game and some of these ways may be nested in one another. We present a model in which agents learn how to perceive the strategic situation they are facing. As a case study we fit the model to data from a laboratory experiment in which participants play a coordination game. Our analysis of the model dynamics gives insights into the costs and benefits of using alternative categorizations of strategies. It also helps us understand which views of the strategy space help to solve the equilibrium selection problem and how agents may learn them through reinforcement.