Reusing Old Policies to Accelerate Learning on New MDPsThe complexity of decentralized control of markov decision processes
