45
18

Contextual Combinatorial Bandits with Probabilistically Triggered Arms

Abstract

We study contextual combinatorial bandits with probabilistically triggered arms (C2^2MAB-T) under a variety of smoothness conditions that capture a wide range of applications, such as contextual cascading bandits and contextual influence maximization bandits. Under the triggering probability modulated (TPM) condition, we devise the C2^2-UCB-T algorithm and propose a novel analysis that achieves an O~(dKT)\tilde{O}(d\sqrt{KT}) regret bound, removing a potentially exponentially large factor O(1/pmin)O(1/p_{\min}), where dd is the dimension of contexts, pminp_{\min} is the minimum positive probability that any arm can be triggered, and batch-size KK is the maximum number of arms that can be triggered per round. Under the variance modulated (VM) or triggering probability and variance modulated (TPVM) conditions, we propose a new variance-adaptive algorithm VAC2^2-UCB and derive a regret bound O~(dT)\tilde{O}(d\sqrt{T}), which is independent of the batch-size KK. As a valuable by-product, we find our analysis technique and variance-adaptive algorithm can be applied to the CMAB-T and C2^2MAB~setting, improving existing results there as well. We also include experiments that demonstrate the improved performance of our algorithms compared with benchmark algorithms on synthetic and real-world datasets.

View on arXiv
Comments on this paper