Ongoing research project

Improved regret for Stochastic Gradient Bandits

This research note studies regret bounds for stochastic gradient bandits, with particular attention to learning rates and reward gaps. It builds on the Lyapunov analysis of Jia and Ouyang.

How to cite

Dorian Baudry. Improved regret for Stochastic Gradient Bandits. Research note, version 1, 16 November 2026. https://dbaudry.github.io/research-notes/sgb-regret/

Download BibTeX

Versions

Back to policy gradients and stochastic approximation