報(bào)告人：北卡羅來納州立大學(xué) Dr . Yunan Liu
時(shí)間：2024年4月15日11：00-12：00
地點(diǎn)：北京郵電大學(xué)西土城校區(qū)教四-441
報(bào)告摘要：
In this talk, we investigate new ways to apply machine learning methodologies to queueing models with applications to service systems (e.g., call centers and healthcare). Our work will cover three different machine learning paradigms: (i) online learning, (ii) offline learning, and (iii) deep learning. (i) We propose a new online reinforcement learning technique to solve a multi-period pricing and staffing problem in a service queueing system with an unknown demand curve. We develop an algorithm called gradient-based online-learning in queues (GOLiQ) to dynamically adjust the service price p (and service rate μ) so as to maximize cumulative expected revenues (the sales revenue minus the delay penalty) over a given finite time horizon. (ii) We develop a new simulation-based offline learning algorithm that can be used to determine the required staffing function that achieves time-stable performance for a time-varying queue within a finite time. Our new algorithm, called simulation-based offline learning staffing algorithm (SOLSA), organizes the overall learning process into successive cycles each of which consists of two phases: (1) (Exploitation) The decision maker generates relevant queueing data via a decision-aware simulator under a candidate solution, (2) (Exploration) Using the newly collected data, improved staffing plans are prescribed and to be used to configure the simulator in the next cycle. (iii) We develop a new deep learning method, dubbed deep learning in non-Markovian queues (DeepLiNQ), which is an offline supervised learning method that learns the system’s intrinsic characteristics using synthetic training data. In real-time applications, DeepLiNQ is built by a set of neuro networks and can be used to recursively provide estimates for the transient system waiting time performance.
專家簡(jiǎn)介：
劉雨楠，美國北卡羅來納州立大學(xué)工業(yè)與系統(tǒng)工程系副教授。于清華大學(xué)電氣工程系獲得學(xué)士學(xué)位，于哥倫比亞大學(xué)工業(yè)工程與運(yùn)籌部獲碩士和博士學(xué)位。研究興趣包括隨機(jī)模型、應(yīng)用概率、仿真、排隊(duì)論、最優(yōu)控制和強(qiáng)化學(xué)習(xí)等，并將其應(yīng)用于客戶聯(lián)絡(luò)中心（customer contact centers）、醫(yī)療保?。?/span>healthcare）、生產(chǎn)和運(yùn)輸系統(tǒng)中。文章發(fā)表本領(lǐng)域旗艦期刊上，如Operations Research, Production and Operations Management, INFORMS Journal on Computing, IISE Transactions, Naval Research Logistics, Stochastic Systems, European Journal of Operational Research, Queueing Systems。榮獲亞馬遜學(xué)者（Amazon Scholar），與亞馬遜客戶服務(wù)部的全球容量規(guī)劃團(tuán)隊(duì)密切合作。個(gè)人網(wǎng)頁：http://yunanliu.wordpress.ncsu.edu
