Off-policy Distributional Q($λ$): Distributional RL without Importance Sampling

Yunhao Tang, Mark Rowland, Rémi Munos, Bernardo Avila Pires, Will Dabney
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:406-414, 2026.

Abstract

We introduce off-policy distributional Q($\lambda$), a new addition to the family of off-policy distributional evaluation algorithms. Off-policy distributional Q($\lambda$) does not apply importance sampling for off-policy learning, which introduces intriguing interactions with signed measures. Such unique properties distributional Q($\lambda$) from other existing alternatives such as distributional Retrace. We characterize the algorithmic properties of distributional Q($\lambda$) and validate theoretical insights with tabular experiments. We show how distributional Q($\lambda$)-C51, a combination of Q($\lambda$) with the C51 agent, exhibits promising results on deep RL benchmarks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-tang26a, title = { Off-policy Distributional Q($λ$): Distributional RL without Importance Sampling }, author = {Tang, Yunhao and Rowland, Mark and Munos, R{\'e}mi and Pires, Bernardo Avila and Dabney, Will}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {406--414}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/tang26a/tang26a.pdf}, url = {https://proceedings.mlr.press/v300/tang26a.html}, abstract = { We introduce off-policy distributional Q($\lambda$), a new addition to the family of off-policy distributional evaluation algorithms. Off-policy distributional Q($\lambda$) does not apply importance sampling for off-policy learning, which introduces intriguing interactions with signed measures. Such unique properties distributional Q($\lambda$) from other existing alternatives such as distributional Retrace. We characterize the algorithmic properties of distributional Q($\lambda$) and validate theoretical insights with tabular experiments. We show how distributional Q($\lambda$)-C51, a combination of Q($\lambda$) with the C51 agent, exhibits promising results on deep RL benchmarks. } }
Endnote
%0 Conference Paper %T Off-policy Distributional Q($λ$): Distributional RL without Importance Sampling %A Yunhao Tang %A Mark Rowland %A Rémi Munos %A Bernardo Avila Pires %A Will Dabney %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-tang26a %I PMLR %P 406--414 %U https://proceedings.mlr.press/v300/tang26a.html %V 300 %X We introduce off-policy distributional Q($\lambda$), a new addition to the family of off-policy distributional evaluation algorithms. Off-policy distributional Q($\lambda$) does not apply importance sampling for off-policy learning, which introduces intriguing interactions with signed measures. Such unique properties distributional Q($\lambda$) from other existing alternatives such as distributional Retrace. We characterize the algorithmic properties of distributional Q($\lambda$) and validate theoretical insights with tabular experiments. We show how distributional Q($\lambda$)-C51, a combination of Q($\lambda$) with the C51 agent, exhibits promising results on deep RL benchmarks.
APA
Tang, Y., Rowland, M., Munos, R., Pires, B.A. & Dabney, W.. (2026). Off-policy Distributional Q($λ$): Distributional RL without Importance Sampling . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:406-414 Available from https://proceedings.mlr.press/v300/tang26a.html.

Related Material