12:00, 15 Oct 1997, WeH 7220 Generalized Prioritized Sweeping David Andre, UC Berkeley Prioritized sweeping is a model-based reinforcement learning method that attempts to focus an agent's limited computational resources to achieve a good estimate of the value of environment states. The classic account of prioritized seeping uses an explicit, state-based representation of the value, reward, and model parameters. Such a representation is unwieldy for dealing with complex environments and there is growing interest in learning with more compact representations. We claim that classic prioritized sweeping is ill-suited for learning with such representations. To overcome this deficiency, we introduce generalized prioritized sweeping, a principled method for generating representation-specific algorithms for model-based reinforcement learning. We have applied this method to several representations, including state-based models and generalized model approximators (such as Bayesian networks). I will describe preliminary experiments that compare our approach with classical prioritized sweeping. Joint work with Nir Friedman and Ronald Parr.