Catching a Moving Subspace: Low-Rank Bandits Beyond Stationarity
The paper titled 'Catching a Moving Subspace: Low-Rank Bandits Beyond Stationarity' explores the challenges of low-rank bandits in dynamic environments. It presents a new algorithm that adapts to changes in the underlying subspace while maintaining efficiency. The authors demonstrate the effectiveness of their approach through empirical results across various benchmarks.
- ▪The study focuses on piecewise-stationary low-rank linear contextual bandits with scalar feedback.
- ▪The proposed algorithm, SPSC, interleaves isotropic probes with exploitation strategies to manage dynamic regret.
- ▪Empirical results show that SPSC outperforms existing non-stationary and low-rank baselines in specific conditions.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Machine Learning arXiv:2605.20269 (cs) [Submitted on 18 May 2026] Title:Catching a Moving Subspace: Low-Rank Bandits Beyond Stationarity Authors:Hamed Khosravi, Xiaoming Huo View a PDF of the paper titled Catching a Moving Subspace: Low-Rank Bandits Beyond Stationarity, by Hamed Khosravi and 1 other authors View PDF HTML (experimental) Abstract:Many bandit deployments (recommendation, clinical dosing, ad targeting) share two facts prior work handles only in isolation: rewards live on a low-dimensional latent subspace, and that subspace drifts. Stationary low-rank bandits exploit rank but break under subspace change; non-stationary linear bandits adapt to drift but pay ambient rate $\widetilde{O}(d\sqrt{T})$.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
