Oracle-Efficient Combinatorial Semi-Bandits
By: Jung-hun Kim, Milan Vojnović, Min-hwan Oh
Potential Business Impact:
Makes smart choices faster with fewer guesses.
We study the combinatorial semi-bandit problem where an agent selects a subset of base arms and receives individual feedback. While this generalizes the classical multi-armed bandit and has broad applicability, its scalability is limited by the high cost of combinatorial optimization, requiring oracle queries at every round. To tackle this, we propose oracle-efficient frameworks that significantly reduce oracle calls while maintaining tight regret guarantees. For the worst-case linear reward setting, our algorithms achieve $\tilde{O}(\sqrt{T})$ regret using only $O(\log\log T)$ oracle queries. We also propose covariance-adaptive algorithms that leverage noise structure for improved regret, and extend our approach to general (non-linear) rewards. Overall, our methods reduce oracle usage from linear to (doubly) logarithmic in time, with strong theoretical guarantees.
Similar Papers
Multi-Play Combinatorial Semi-Bandit Problem
Machine Learning (CS)
Helps computers make better choices with more options.
Near-Optimal Regret for Efficient Stochastic Combinatorial Semi-Bandits
Machine Learning (CS)
Helps computers pick best options faster.
Algorithm Design and Stronger Guarantees for the Improving Multi-Armed Bandits Problem
Machine Learning (CS)
Helps computers pick the best option faster.