Direct LCB-based timing-bandit algorithm

Determine how a direct lower-confidence-bound algorithm can efficiently exploit the consecutive feedback structure of the Timing Bandit problem while preserving favorable regret guarantees.

Background

The paper introduces OE-BCAE, which combines lower-confidence-bound selection with the elimination-based BCAE framework. The authors note that a purely LCB-based design would be more direct but that its interaction with consecutive feedback is not understood.

The unresolved issue is both algorithmic and theoretical: it is unclear how to design a direct LCB variant that uses consecutive observations efficiently and achieves a regret bound comparable to the paper’s structure-aware methods.