Identifying Useful Subgoals in Reinforcement Learning by Local Graph Partitioning

Abstract

We present a new subgoal-based method for automatically creating useful skills in reinforcement learning. Our method identifies subgoals by partitioning local state transition graphsâthose that are constructed using only the most recent experiences of the agent. The local scope of our subgoal discovery method allows it to successfully identify the type of subgoals we seekâstates that lie between two densely-connected regions of the state spaceâwhile producing an algorithm with low computational cost.

Cite

Text

Simsek et al. "Identifying Useful Subgoals in Reinforcement Learning by Local Graph Partitioning." International Conference on Machine Learning, 2005. doi:10.1145/1102351.1102454

Markdown

[Simsek et al. "Identifying Useful Subgoals in Reinforcement Learning by Local Graph Partitioning." International Conference on Machine Learning, 2005.](https://mlanthology.org/icml/2005/simsek2005icml-identifying/) doi:10.1145/1102351.1102454

BibTeX

@inproceedings{simsek2005icml-identifying,
  title     = {{Identifying Useful Subgoals in Reinforcement Learning by Local Graph Partitioning}},
  author    = {Simsek, Özgür and Wolfe, Alicia P. and Barto, Andrew G.},
  booktitle = {International Conference on Machine Learning},
  year      = {2005},
  pages     = {816-823},
  doi       = {10.1145/1102351.1102454},
  url       = {https://mlanthology.org/icml/2005/simsek2005icml-identifying/}
}