Develops a directional bias certificate (M_bias, ρ) and Ellipsoidal-MINUCB algorithm that assigns direction-specific bias budgets, enabling improved regret in aligned offline-to-online linear bandit settings.
Since only pulls of armj ⋆ distinguishP 0 fromP j⋆, KL(P0, Pj⋆) =E 0[Nj⋆]·kl 1 4 , 1 4 + ρ√ 2 ≤ 4ρ2T d−1 ≤ 1 16
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Direction-Aware Offline-to-Online Learning in Linear Contextual Bandits
Develops a directional bias certificate (M_bias, ρ) and Ellipsoidal-MINUCB algorithm that assigns direction-specific bias budgets, enabling improved regret in aligned offline-to-online linear bandit settings.