Effective sketching methods for value function approximation

Yangchen Pan, Erfan Sadeqi Azer, Martha White
Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence, PMLR R15:301-310, 2017.

Abstract

High-dimensional representations, such as ra- dial basis function networks or tile cod- ing, are common choices for policy evalua- tion in reinforcement learning. Learning with such high-dimensional representations, how- ever, can be expensive, particularly for matrix methods, such as least-squares temporal differ- ence learning or quasi-Newton methods that approximate matrix step-sizes. In this work, we explore the utility of sketching for these two classes of algorithms. We highlight issues with sketching the high-dimensional features directly, which can incur significant bias. As a remedy, we demonstrate how to use sketching more sparingly, with only a left-sided sketch, that can still enable significant computational gains and the use of these matrix-based learn- ing algorithms that are less sensitive to param- eters. We empirically investigate these algo- rithms, in four domains with a variety of repre- sentations. Our aim is to provide insights into effective use of sketching in practice.

Cite this Paper

BibTeX
@InProceedings{pmlr-vR15-pan17a, title = {Effective sketching methods for value function approximation}, author = {Pan, Yangchen and Azer, Erfan Sadeqi and White, Martha}, booktitle = {Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence}, pages = {301--310}, year = {2017}, editor = {Elidan, Gal and Kersting, Kristian}, volume = {R15}, series = {Proceedings of Machine Learning Research}, month = {11--15 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/r15/main/assets/pan17a/pan17a.pdf}, url = {https://proceedings.mlr.press/r15/pan17a.html}, abstract = {High-dimensional representations, such as ra- dial basis function networks or tile cod- ing, are common choices for policy evalua- tion in reinforcement learning. Learning with such high-dimensional representations, how- ever, can be expensive, particularly for matrix methods, such as least-squares temporal differ- ence learning or quasi-Newton methods that approximate matrix step-sizes. In this work, we explore the utility of sketching for these two classes of algorithms. We highlight issues with sketching the high-dimensional features directly, which can incur significant bias. As a remedy, we demonstrate how to use sketching more sparingly, with only a left-sided sketch, that can still enable significant computational gains and the use of these matrix-based learn- ing algorithms that are less sensitive to param- eters. We empirically investigate these algo- rithms, in four domains with a variety of repre- sentations. Our aim is to provide insights into effective use of sketching in practice.}, note = {Reissued by PMLR on 04 October 2026.} }
Endnote
%0 Conference Paper %T Effective sketching methods for value function approximation %A Yangchen Pan %A Erfan Sadeqi Azer %A Martha White %B Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2017 %E Gal Elidan %E Kristian Kersting %F pmlr-vR15-pan17a %I PMLR %P 301--310 %U https://proceedings.mlr.press/r15/pan17a.html %V R15 %X High-dimensional representations, such as ra- dial basis function networks or tile cod- ing, are common choices for policy evalua- tion in reinforcement learning. Learning with such high-dimensional representations, how- ever, can be expensive, particularly for matrix methods, such as least-squares temporal differ- ence learning or quasi-Newton methods that approximate matrix step-sizes. In this work, we explore the utility of sketching for these two classes of algorithms. We highlight issues with sketching the high-dimensional features directly, which can incur significant bias. As a remedy, we demonstrate how to use sketching more sparingly, with only a left-sided sketch, that can still enable significant computational gains and the use of these matrix-based learn- ing algorithms that are less sensitive to param- eters. We empirically investigate these algo- rithms, in four domains with a variety of repre- sentations. Our aim is to provide insights into effective use of sketching in practice. %Z Reissued by PMLR on 04 October 2026.
APA
Pan, Y., Azer, E.S. & White, M.. (2017). Effective sketching methods for value function approximation. Proceedings of the 33rd Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research R15:301-310 Available from https://proceedings.mlr.press/r15/pan17a.html. Reissued by PMLR on 04 October 2026.

Related Material