Learning-Aided Control in Stochastic Queueing Systems

In this paper, we investigate the power of online learningin stochastic network optimization with unknown systemstatistics a prior. We are interested in understanding howinformation and learning can be efficiently incorporated intosystem control techniques, and what are the fundamentalbenefits of doing so. We propose two Online Learning-AidedControl techniques, OLAC and OLAC2, that explicitly utilizethe past system information in current system control viaa learning procedure called dual learning. We prove strongperformance guarantees of the proposed algorithms: OLACand OLAC2 achieve the near-optimal [O( ), O([log(1/ )])]utility-delay tradeoff and OLAC2 possesses an O( −2/3) con-vergence time. OLAC and OLAC2 are probably the first al-gorithms that simultaneously possess explicit near-optimaldelay guarantee and sub-linear convergence time. Simula-tion results also confirm the superior performance of theproposed algorithms in practice. To the best of our knowl-edge, our attempt is the first to explicitly incorporate onlinelearning into stochastic network optimization and to demon-strate its power in both theory and practice.