Warm-Start UCB
An online learning algorithm for delayed one-sided feedback settings that combines offline warm-start initialization with a UCB exploration strategy. It is applicable to contextual decision problems such as personalized memory gating.
Timeline (last 90 days)
The Warm-Start UCB algorithm first uses existing offline data to learn an initial policy, then continuously adjusts it during online interaction with the current user.
Theoretical analysis shows that when the delay window is no more than O(√T), the high-probability regret upper bound of the Warm-Start UCB algorithm is Õ(Tε ∧ d√T), where T is the total number of rounds, ε is the initialization error, and d is the feature dimension.
Experimental results show that Warm-Start UCB outperforms all baseline methods under different memory budgets and history lengths.
Personalized gating can improve memory utilization efficiency, and the combination of offline warm start plus online adaptation is effective.
All Facts (4)
The Warm-Start UCB algorithm first uses existing offline data to learn an initial policy, then continuously adjusts it during online interaction with the current user.
50%UnverifiedTheoretical analysis shows that when the delay window is no more than O(√T), the high-probability regret upper bound of the Warm-Start UCB algorithm is Õ(Tε ∧ d√T), where T is the total number of rounds, ε is the initialization error, and d is the feature dimension.
50%UnverifiedExperimental results show that Warm-Start UCB outperforms all baseline methods under different memory budgets and history lengths.
50%UnverifiedPersonalized gating can improve memory utilization efficiency, and the combination of offline warm start plus online adaptation is effective.
50%