English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The problem of online learning with graph feedback has been extensivelystudied in the literature due to its generality and potential to model variouslearning tasks. Existing works mainly study the adversarial and stochasticfeedback separately. If the prior knowledge of the feedback mechanism isunavailable or wrong, such specially designed algorithms could suffer greatloss. To avoid this problem, \citet{erez2021towards} try to optimize for bothenvironments. However, they assume the feedback graphs are undirected and eachvertex has a self-loop, which compromises the generality of the framework andmay not be satisfied in applications. With a general feedback graph, theobservation of an arm may not be available when this arm is pulled, which makesthe exploration more expensive and the algorithms more challenging to performoptimally in both environments. In this work, we overcome this difficulty by anew trade-off mechanism with a carefully-designed proportion for explorationand exploitation. We prove the proposed algorithm simultaneously achieves$\mathrm{poly} \log T$ regret in the stochastic setting and minimax-optimalregret of $\tilde{O}(T^{2/3})$ in the adversarial setting where $T$ is thehorizon and $\tilde{O}$ hides parameters independent of $T$ as well aslogarithmic terms. To our knowledge, this is the first best-of-both-worldsresult for general feedback graphs.

Simultaneously Learning Stochastic and Adversarial Bandits with General Graph Feedback