摘要：The area of intelligent systems is rapidly developing and reaching in multiple directions. In this mini-special issue we present four papers that show some of the area that current intelligent systems enclose. The first paper, entitled “The Cross-Entropy Method for policy search in Decentralized POMDPs,” was written by Frans A. Oliehoek, Julian F.P. Kooij, and Nikos Vlassis. It concerns possible approaches to multiagent planning under uncertainty. One of possible methods is utilization of Decentralized Partially Observable Markov Decision Processes (Dec-POMDPs). Unfortunately, solving a Dec-POMDP exactly is an intractable combinatorial optimization problem. In the paper, a new Cross-Entropy-based method is applied to the problem. It operates by sampling pure policies from an appropriately parameterized stochastic policy, and then evaluates them to define the next stochastic policy to sample from. This process is repeated until convergence is reached. Experimental results demonstrate that the proposed method can efficiently search large solution spaces.
