Résumé

Players may categorize the strategies available to them. In many games there are different ways to categorize one's strategies (different frames). This paper presents a model of agents who learn which internal representation of the strategy space to use. We fit the model to existing experimental data and show that the model incorporating frames explains the data better than a basic reinforcement learning model. The analysis identifies trade-offs of using coarse versus fine frames and coarse versus fine categories of strategies when it comes to learning.

Référence

Vessela Daskalova, « Learning Frames », IAST Lunch Seminar, Toulouse : IAST, 19 mars 2019, 11h45–12h45, salle MS001.