Autoregressive Bandits

Francesco Bacchiocchi, Gianmarco Genalti, Davide Maran, Marco Mussi, Marcello Restelli, Nicola Gatti, Alberto Maria Metelli
Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, PMLR 238:937-945, 2024.

Abstract

Autoregressive processes naturally arise in a large variety of real-world scenarios, including stock markets, sales forecasting, weather prediction, advertising, and pricing. When facing a sequential decision-making problem in such a context, the temporal dependence between consecutive observations should be properly accounted for guaranteeing convergence to the optimal policy. In this work, we propose a novel online learning setting, namely, Autoregressive Bandits (ARBs), in which the observed reward is governed by an autoregressive process of order $k$, whose parameters depend on the chosen action. We show that, under mild assumptions on the reward process, the optimal policy can be conveniently computed. Then, we devise a new optimistic regret minimization algorithm, namely, AutoRegressive Upper Confidence Bound (AR-UCB), that suffers sublinear regret of order $\tilde{O} ( \frac{(k+1)^{3/2}\sqrt{nT}}{(1-\Gamma)^2} )$, where $T$ is the optimization horizon, $n$ is the number of actions, and $\Gamma < 1$ is a stability index of the process. Finally, we empirically validate our algorithm, illustrating its advantages w.r.t. bandit baselines and its robustness to misspecification of key parameters.

Cite this Paper

BibTeX
@InProceedings{pmlr-v238-bacchiocchi24a, title = { Autoregressive Bandits }, author = {Bacchiocchi, Francesco and Genalti, Gianmarco and Maran, Davide and Mussi, Marco and Restelli, Marcello and Gatti, Nicola and Maria Metelli, Alberto}, booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics}, pages = {937--945}, year = {2024}, editor = {Dasgupta, Sanjoy and Mandt, Stephan and Li, Yingzhen}, volume = {238}, series = {Proceedings of Machine Learning Research}, month = {02--04 May}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v238/bacchiocchi24a/bacchiocchi24a.pdf}, url = {https://proceedings.mlr.press/v238/bacchiocchi24a.html}, abstract = { Autoregressive processes naturally arise in a large variety of real-world scenarios, including stock markets, sales forecasting, weather prediction, advertising, and pricing. When facing a sequential decision-making problem in such a context, the temporal dependence between consecutive observations should be properly accounted for guaranteeing convergence to the optimal policy. In this work, we propose a novel online learning setting, namely, Autoregressive Bandits (ARBs), in which the observed reward is governed by an autoregressive process of order $k$, whose parameters depend on the chosen action. We show that, under mild assumptions on the reward process, the optimal policy can be conveniently computed. Then, we devise a new optimistic regret minimization algorithm, namely, AutoRegressive Upper Confidence Bound (AR-UCB), that suffers sublinear regret of order $\tilde{O} ( \frac{(k+1)^{3/2}\sqrt{nT}}{(1-\Gamma)^2} )$, where $T$ is the optimization horizon, $n$ is the number of actions, and $\Gamma < 1$ is a stability index of the process. Finally, we empirically validate our algorithm, illustrating its advantages w.r.t. bandit baselines and its robustness to misspecification of key parameters. } }
Endnote
%0 Conference Paper %T Autoregressive Bandits %A Francesco Bacchiocchi %A Gianmarco Genalti %A Davide Maran %A Marco Mussi %A Marcello Restelli %A Nicola Gatti %A Alberto Maria Metelli %B Proceedings of The 27th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2024 %E Sanjoy Dasgupta %E Stephan Mandt %E Yingzhen Li %F pmlr-v238-bacchiocchi24a %I PMLR %P 937--945 %U https://proceedings.mlr.press/v238/bacchiocchi24a.html %V 238 %X Autoregressive processes naturally arise in a large variety of real-world scenarios, including stock markets, sales forecasting, weather prediction, advertising, and pricing. When facing a sequential decision-making problem in such a context, the temporal dependence between consecutive observations should be properly accounted for guaranteeing convergence to the optimal policy. In this work, we propose a novel online learning setting, namely, Autoregressive Bandits (ARBs), in which the observed reward is governed by an autoregressive process of order $k$, whose parameters depend on the chosen action. We show that, under mild assumptions on the reward process, the optimal policy can be conveniently computed. Then, we devise a new optimistic regret minimization algorithm, namely, AutoRegressive Upper Confidence Bound (AR-UCB), that suffers sublinear regret of order $\tilde{O} ( \frac{(k+1)^{3/2}\sqrt{nT}}{(1-\Gamma)^2} )$, where $T$ is the optimization horizon, $n$ is the number of actions, and $\Gamma < 1$ is a stability index of the process. Finally, we empirically validate our algorithm, illustrating its advantages w.r.t. bandit baselines and its robustness to misspecification of key parameters.
APA
Bacchiocchi, F., Genalti, G., Maran, D., Mussi, M., Restelli, M., Gatti, N. & Maria Metelli, A.. (2024). Autoregressive Bandits . Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 238:937-945 Available from https://proceedings.mlr.press/v238/bacchiocchi24a.html.

Related Material