机器学习与数据科学博士生系列论坛（第五十九期）—— Gradient Based Online Learning for Queueing Systems: A Gentle Introduction

摘要：
Queueing systems find diverse applications across various domains. They are commonly used in telecommunications, transportation, computer resource allocation, and healthcare settings. These applications highlight the significance of queueing systems in optimizing processes and resource utilization. Within these systems, we often aim to achieve specific objectives by adjusting parameters, such as maximizing our net profit, minimizing the mean steady-state system time, or identifying Nash equilibrium points in queueing games. In highly streaming and non-iid data environments like queueing systems, our primary concern is how to efficiently and promptly identify target parameters. To tackle these issues, (stochastic) gradient descent is the top choice.

In this talk, we will explore classical queueing model problems, along with several mainstream algorithms for addressing them. These algorithms are all adaptations of SGD. Specially, we will focus on the dynamic pricing and capacity sizing problem in a G/G/1 service system, emphasizing the application of the SAMCMC algorithm. Leveraging SAMCMC, we can achieve the quickest statistical estimation of target parameters, minimizing our cumulative regret. Ultimately, we will explore how SAMCMC enables online statistical inference for target parameters in queueing systems, a capability that previous methods could not offer.

论坛简介：该线上论坛是由张志华教授机器学习实验室组织，每两周主办一次（除了公共假期）。论坛每次邀请一位博士生就某个前沿课题做较为系统深入的介绍，主题包括但不限于机器学习、高维统计学、运筹优化和理论计算机科学。