Learning to identify winning coalitions in the PAC model

We consider PAC learning of simple cooperative games, in which the coalitions are partitioned into "winning" and "losing" coalitions. We analyze the complexity of learning a suitable concept class via its Vapnik-Chervonenkis (VC) dimension, and provide an algorithm that learns this class. Furthermore, we study constrained simple games; we demonstrate that the VC dimension can be dramatically reduced when one allows only a single minimum winning coalition (even more so when this coalition has cardinality 1), whereas other interesting constraints do not significantly lower the dimension. Categories and Subject Descriptors F.2 [Theory of Computation]: Analysis of Algorithms and Problem Complexity; I.2.6 [Artificial Intelligence]: Learning--Concept Learning; I.2.11 [Artificial Intelligence]: Distributed Artificial Intelligence--Multiagent Systems General Terms Algorithms, Theory Keywords PAC Learning, Coalition Formation

• Type: Conference
• Year: 2006
• Where: ATAL
• Authors: Ariel D. Procaccia, Jeffrey S. Rosenschein
• Keywords: ATAL 2006, Coalition, Intelligent Agents, Pac Learning, Simple Cooperative Games