February 23, 2024Open Access

Towards Efficient and Optimal Covariance-Adaptive Algorithms for Combinatorial Semi-Bandits

View Full Paper

Ask AI

Bookmark

Authors

Discussion

Randomized trial shows improved regret in semi-bandit problems, suggesting better decision-making approaches.

Improved regret is observed with covariance-adaptive algorithms that estimate the covariance matrix online, demonstrating effective learning and performance.
The two algorithms OLSUCBC and COSV yield better gap-free regret results in various scenarios, enhancing practical applications of semi-bandit strategies.
Utilizing covariance structures leads to significant performance enhancements compared to traditional bandit feedback methods across different settings and conditions for decision-making tasks in combinatorial settings with items and actions involved. Enhanced computational efficiency is achieved in COSV by drawing inspiration from Thompson Sampling techniques, illustrating possible trade-offs in algorithm design.

Zhou et al. (2024) studied this question.

View Full Paper

Ask AI

Bookmark