Belangrijkste concepten
Proposing the α-Linear-Contextual (LC)-Tsallis-INF algorithm for linear contextual bandits with improved regret bounds.
Samenvatting
This study introduces the α-Linear-Contextual (LC)-Tsallis-INF algorithm for linear contextual bandits. It addresses the linear contextual bandit problem with adversarial corruption, proposing a Best-of-Both-Worlds algorithm. The content discusses the existing studies, assumptions, and the proposed algorithm's structure and parameters. It provides detailed insights into the linear contextual bandit problem, the proposed algorithm, and the theoretical analysis of regret bounds.
Statistieken
O(log2(T )) regret upper bound in existing studies.
O(log(T )) regret upper bound using the reduction framework.
O(log(T )) regret upper bound with the Tsallis entropy.
Citaten
"We introduce a margin condition to characterize the problem difficulty."
"Our proposed algorithm is based on the Follow-The-Regularized-Leader with the Tsallis entropy."