摘要：

Most agent-based modeling techniques generate only a single trajectory in each run, greatly undersampling the space of possible trajectories. Swarming agents can explore many alternative futures in parallel, particularly when they interact through digital pheromone fields. This paper shows how these fields and other artifacts developed by such a model can be interpreted as conditional probabilities estimated by sampling a very large number of possible trajectories. This interpretation offers several benefits. It supports theoretical insight into the behavior of swarming models by mapping them onto more traditional probabilistic models such as Markov decision processes, it allows us to derive more information from them than swarming models usually yield, and it facilitates integrating them with probability-based AI mechanisms such as HMM's or Bayesian networks.

展开