ConceptKL-UCB算法
KL-UCB
KL-UCB is a multi-armed bandit algorithm that constructs upper confidence bounds based on KL divergence. Under Bernoulli reward distributions it achieves tighter regret bounds than algorithms such as UCB1, making it one of the classic optimal bandit algorithms.
Timeline (last 90 days)
Oct 16
The KL-UCB algorithm constructs confidence upper bounds based on KL divergence and, compared with algorithms such as UCB1 that are based on Hoeffding's inequality, can achieve tighter regret bounds under Bernoulli rewards.
Unverified50%