Double Doubly Robust Thompson Sampling for Generalized Linear Contextual Bandits

We propose a novel algorithm for generalized linear contextual bandits (GLBs) with regret bound sublinear to the time horizon, minimum eigenvalue of covariance contexts and lower variance rewards. In several identified cases, our result is first achieving dimension without discarding observed Previous approaches achieve by rewards, whereas achieves incorporating from all arms in double doubly robust (DDR) estimator. The DDR estimator subclass but tighter error bound. also provide logarithmic cumulative under probabilistic margin condition. This condition models or GLMs when are different coefficients common. conduct empirical studies using synthetic data real examples, demonstrating effectiveness algorithm.