Non-Parametric Approximate Dynamic Programming via the Kernel Method

Abstract

This paper presents a novel non-parametric approximate dynamic programming (ADP) algorithm that enjoys graceful, dimension-independent approximation and sample complexity guarantees. In particular, we establish both theoretically and computationally that our proposal can serve as a viable alternative to state-of-the-art parametric ADP algorithms, freeing the designer from carefully specifying an approximation architecture. We accomplish this by developing a kernel-based mathematical program for ADP. Via a computational study on a controlled queueing network, we show that our non-parametric procedure is competitive with parametric ADP approaches.

Cite

Text

Bhat et al. "Non-Parametric Approximate Dynamic Programming via the Kernel Method." Neural Information Processing Systems, 2012.

Markdown

[Bhat et al. "Non-Parametric Approximate Dynamic Programming via the Kernel Method." Neural Information Processing Systems, 2012.](https://mlanthology.org/neurips/2012/bhat2012neurips-nonparametric/)

BibTeX

@inproceedings{bhat2012neurips-nonparametric,
  title     = {{Non-Parametric Approximate Dynamic Programming via the Kernel Method}},
  author    = {Bhat, Nikhil and Farias, Vivek and Moallemi, Ciamac C.},
  booktitle = {Neural Information Processing Systems},
  year      = {2012},
  pages     = {386-394},
  url       = {https://mlanthology.org/neurips/2012/bhat2012neurips-nonparametric/}
}