Extensions beyond common priorities and one-sided uncertainty

Investigate and establish extensions of exact regret-frontier characterization and externality scheduling for centralized serial-dictatorship bandits beyond a known common priority order and under two-sided preference uncertainty.

Background

The paper's results assume a known common priority order and one-sided uncertainty in player-arm reward means. Under these assumptions, pairwise exploration quotas, the compact marginal linear program, the attainable regret region, and frontier-attaining policies are characterized.

The conclusion explicitly identifies relaxing the common-priority assumption and treating two-sided uncertainty as unresolved directions for extending the framework.