[2311.08376] Ensemble sampling for linear bandits: small ensembles suffice


Obtain a PDF of the paper titled Ensemble sampling for linear bandits: small ensembles suffice, by David Janz and a couple of different authors

Obtain PDF
HTML (experimental)

Summary:We offer the primary helpful and rigorous evaluation of ensemble sampling for the stochastic linear bandit setting. Particularly, we present that, beneath customary assumptions, for a $d$-dimensional stochastic linear bandit with an interplay horizon $T$, ensemble sampling with an ensemble of dimension of order $smash{d log T}$ incurs remorse at many of the order $smash{(d log T)^{5/2} sqrt{T}}$. Ours is the primary lead to any structured setting to not require the dimensions of the ensemble to scale linearly with $T$ — which defeats the aim of ensemble sampling — whereas acquiring close to $smash{sqrt{T}}$ order remorse. Ours can be the primary outcome that permits infinite motion units.

Submission historical past

From: David Janz [view email]
Tue, 14 Nov 2023 18:41:28 UTC (36 KB)
Wed, 6 Mar 2024 12:31:27 UTC (45 KB)

Supply hyperlink


Please enter your comment!
Please enter your name here