Robust Tests in Online Decision-Making

Bandit algorithms are widely used in sequential decision problems to maximize the cumulative reward. One potential application is mobile health, where goal promote user's health through personalized interventions based on user specific information acquired wearable devices. Important considerations include type of, and frequency with which data collected (e.g. GPS, or continuous monitoring), as such factors can severely impact app performance users’ adherence. In order balance need collect that useful constraint of impacting performance, one needs be able assess usefulness variables. feedback sequentially correlated, so traditional testing procedures developed for independent cannot apply. Recently, a statistical procedure was actor-critic bandit algorithm. An algorithm maintains two separate models, actor, action selection policy, other critic, reward model. The well validity test guaranteed only when critic model correctly specified. However, misspecification frequent practice due incorrect functional form missing covariates. this work, we propose modified robust derive novel actor parameters case.