Combinatorial Pure Exploration

Combinatorial pure exploration focuses on efficiently identifying the best subset of "arms" (e.g., features, actions) from a large set to maximize a reward, given limited exploration budget. Recent research emphasizes developing algorithms, such as Thompson Sampling and successive elimination methods, that achieve optimal or near-optimal sample complexity even when the number of possible subsets is exponentially large. This work addresses both theoretical guarantees on performance and practical computational efficiency, particularly for large-scale problems, with applications ranging from clinical trials to online advertising. The development of efficient and provably optimal algorithms is crucial for advancing the field and enabling real-world applications of this important online learning problem.