Neural Variance-aware Dueling Bandits with Deep Representation and Shallow Exploration

Youngmin Oh, Jinje Park, Taejin Paik
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:550-558, 2026.

Abstract

We introduce the first variance-aware algorithms for contextual dueling bandits that leverage shallow exploration strategies with neural networks for nonlinear utility approximation. A key theoretical challenge is the absence of a closed-form estimator, which led prior work to require an extremely large network width $m$ (i.e., $m = \widetilde{\Omega}(T^{14})$). We address this constraint with a novel analytical approach that combines iterative self-improvement with spectral analysis. Our analysis significantly reduces the network width requirement to $m = \widetilde{\Omega}(T^{6})$, and shows that our algorithms achieve a sublinear regret of $ \widetilde{\mathcal{O}}\left(d\sqrt{\sum_{t=1}^{T} \sigma_t^2} + \sqrt{dT}\right) $ under both UCB and TS frameworks. Empirical results show that the proposed algorithms are not only computationally efficient and exhibit sublinear regret in practical settings, but also achieve state-of-the-art performance on both synthetic and real-world tasks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-oh26a, title = { Neural Variance-aware Dueling Bandits with Deep Representation and Shallow Exploration }, author = {Oh, Youngmin and Park, Jinje and Paik, Taejin}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {550--558}, 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/oh26a/oh26a.pdf}, url = {https://proceedings.mlr.press/v300/oh26a.html}, abstract = { We introduce the first variance-aware algorithms for contextual dueling bandits that leverage shallow exploration strategies with neural networks for nonlinear utility approximation. A key theoretical challenge is the absence of a closed-form estimator, which led prior work to require an extremely large network width $m$ (i.e., $m = \widetilde{\Omega}(T^{14})$). We address this constraint with a novel analytical approach that combines iterative self-improvement with spectral analysis. Our analysis significantly reduces the network width requirement to $m = \widetilde{\Omega}(T^{6})$, and shows that our algorithms achieve a sublinear regret of $ \widetilde{\mathcal{O}}\left(d\sqrt{\sum_{t=1}^{T} \sigma_t^2} + \sqrt{dT}\right) $ under both UCB and TS frameworks. Empirical results show that the proposed algorithms are not only computationally efficient and exhibit sublinear regret in practical settings, but also achieve state-of-the-art performance on both synthetic and real-world tasks. } }
Endnote
%0 Conference Paper %T Neural Variance-aware Dueling Bandits with Deep Representation and Shallow Exploration %A Youngmin Oh %A Jinje Park %A Taejin Paik %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-oh26a %I PMLR %P 550--558 %U https://proceedings.mlr.press/v300/oh26a.html %V 300 %X We introduce the first variance-aware algorithms for contextual dueling bandits that leverage shallow exploration strategies with neural networks for nonlinear utility approximation. A key theoretical challenge is the absence of a closed-form estimator, which led prior work to require an extremely large network width $m$ (i.e., $m = \widetilde{\Omega}(T^{14})$). We address this constraint with a novel analytical approach that combines iterative self-improvement with spectral analysis. Our analysis significantly reduces the network width requirement to $m = \widetilde{\Omega}(T^{6})$, and shows that our algorithms achieve a sublinear regret of $ \widetilde{\mathcal{O}}\left(d\sqrt{\sum_{t=1}^{T} \sigma_t^2} + \sqrt{dT}\right) $ under both UCB and TS frameworks. Empirical results show that the proposed algorithms are not only computationally efficient and exhibit sublinear regret in practical settings, but also achieve state-of-the-art performance on both synthetic and real-world tasks.
APA
Oh, Y., Park, J. & Paik, T.. (2026). Neural Variance-aware Dueling Bandits with Deep Representation and Shallow Exploration . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:550-558 Available from https://proceedings.mlr.press/v300/oh26a.html.

Related Material