Skip to yearly menu bar Skip to main content


Best-of-Both-Worlds Algorithms for Linear Contextual Bandits

Yuko Kuroki · Alberto Rumi · Taira Tsuchiya · Fabio Vitale · Nicolò Cesa-Bianchi

MR1 & MR2 - Number 101
[ ]
Fri 3 May 8 a.m. PDT — 8:30 a.m. PDT
Oral presentation: Oral: Bandit & Causality
Fri 3 May 1:30 a.m. PDT — 2:30 a.m. PDT

Abstract: We study best-of-both-worlds algorithms for $K$-armed linear contextual bandits. Our algorithms deliver near-optimal regret bounds in both the adversarial and stochastic regimes, without prior knowledge about the environment. In the stochastic regime, we achieve the polylogarithmic rate $\frac{(dK)^2\poly\!\log(dKT)}{\Delta_{\min}}$, where $\Delta_{\min}$ is the mininum suboptimality gap over the $d$-dimensional context space. In the adversarial regime, we obtain either the first-order $\widetilde{\cO}(dK\sqrt{L^*})$ bound, or the second-order $\widetilde{\cO}(dK\sqrt{\Lambda^*})$ bound, where $L^*$ is the cumulative loss of the best action and $\Lambda^*$ is a notion of the cumulative second moment for the losses incurred by the algorithm. Moreover, we develop an algorithm based on FTRL with Shannon entropy regularizer that does not require the knowledge of the inverse of the covariance matrix, and achieves a polylogarithmic regret in the stochastic regime while obtaining $\widetilde\cO\big(dK\sqrt{T}\big)$ regret bounds in the adversarial regime.

Live content is unavailable. Log in and register to view live content