The Bayesian Prophet: A Low-Regret Framework for Online Decision Making

We develop a new framework for designing online policies given access to an oracle providing statistical information about off-line benchmark. Having such prediction oracles enables simple and natural Bayesian selection raises the question as how these perform in different settings. Our work makes two important contributions toward this question: First, we general technique call compensated coupling, which can be used derive bounds on expected regret (i.e., additive loss with respect benchmark) any policy Second, using technique, show that greedy policy, Bayes selector, has constant independent of number arrivals resource levels) large class problems refer “online allocation finite types,” includes widely studied packing matching problems. results generalize simplify several existing suggest promising pathway obtaining oracle-driven other decision-making This paper was accepted by George Shanthikumar, big data analytics.