Efficient Hierarchical MCMC for Policy Search

Abstract

Many inference and optimization tasks in machine learning can be solved bysampling approaches such as Markov Chain Monte Carlo (MCMC) and simulatedannealing. These methods can be slow if a single target density query requiresmany runs of a simulation (or a complete sweep of a training data set). Weintroduce a hierarchy of MCMC samplers that allow most steps to be taken inthe solution space using only a small sample of simulation runs (or trainingexamples). This is shown to accelerate learning in a policy searchoptimization task.

Cite

Text

Strens. "Efficient Hierarchical MCMC for Policy Search." International Conference on Machine Learning, 2004. doi:10.1145/1015330.1015381

Markdown

[Strens. "Efficient Hierarchical MCMC for Policy Search." International Conference on Machine Learning, 2004.](https://mlanthology.org/icml/2004/strens2004icml-efficient/) doi:10.1145/1015330.1015381

BibTeX

@inproceedings{strens2004icml-efficient,
  title     = {{Efficient Hierarchical MCMC for Policy Search}},
  author    = {Strens, Malcolm J. A.},
  booktitle = {International Conference on Machine Learning},
  year      = {2004},
  doi       = {10.1145/1015330.1015381},
  url       = {https://mlanthology.org/icml/2004/strens2004icml-efficient/}
}