Reviews: Fully Decentralized Policies for Multi-Agent Systems: An Information Theoretic Approach

Neural Information Processing Systems 

There is also a large body of work on distributed and decentralized methods from systems and control theory, which should be mentioned in this context (see e.g.