Structured Prioritized Sweeping

@inproceedings{DeardenStructuredPS,
  title={Structured Prioritized Sweeping},
  author={Richard Dearden}
}
The structured policy iteration (SPI) algorithm (Boutilier et al., 1995) constructs structured solutions to Markov decision problems (MDPs). It finds the optimal value function by treating all states which have the same value under a particular policy as a single aggregate state. Here we show that this approach can also be applied locally in a structured version of prioritized sweeping, a model-based reinforcement learning algorithm that attempts to focus the learning agent’s limited… CONTINUE READING
8 Citations
16 References
Similar Papers

Similar Papers

Loading similar papers…