講座題目：基于機(jī)器學(xué)習(xí)和金融風(fēng)險(xiǎn)控制的供應(yīng)商采購(gòu)決策

報(bào) 告 人：Youhua (Frank) Chen

時(shí) 　 間：2019年12月27日(周五)14:30-16:30

地 　 點(diǎn)：中關(guān)村校區(qū)主樓317室

主辦單位：研究生院,、管理與經(jīng)濟(jì)學(xué)院

報(bào)名方式：登錄北京理工大學(xué)微信企業(yè)號(hào)---第二課堂---課程報(bào)名中選擇“【百家大講堂】第306期：基于機(jī)器學(xué)習(xí)和金融風(fēng)險(xiǎn)控制的供應(yīng)商采購(gòu)決策”

【主講人簡(jiǎn)介】

  Youhua (Frank) Chen,，多倫多大學(xué)博士，現(xiàn)任香港城市大學(xué)管理科學(xué)系講座教授及系主任,。在2012年加入香港城市大學(xué)之前,，Youhua (Frank) Chen教授曾在新加坡國(guó)立大學(xué)商學(xué)院（1997-2001）和香港中文大學(xué)系統(tǒng)工程與工程管理系（2001-2012）任職,。Youhua (Frank) Chen教授的研究興趣包括共享經(jīng)濟(jì)、醫(yī)療健康管理,、供應(yīng)鏈建模和庫(kù)存系統(tǒng)分析,在OR,、MS、POM,、M&SOM,、NRL等運(yùn)作管理領(lǐng)域國(guó)際頂級(jí)期刊發(fā)表多篇學(xué)術(shù)論文，例如代表作“Quantifying the bullwhip effect in a simple supply chain: The impact of forecasting, lead times, and information”發(fā)表后已經(jīng)被引用2200余篇次,，在供應(yīng)鏈管理領(lǐng)域名列前茅,。

Prof. Youhua (Frank) Chen is Chair Professor and Head of Management Sciences at City University of Hong Kong. He holds a bachelor’s degree in Engineering, master’s degree in Economics, and doctoral degree in Management from Tsinghua University, the University of Waterloo, and the University of Toronto, respectively. Before joining National University of Singapore in 1997, he took a post-doctoral fellow position at Northwestern University. After 11 years of teaching at the Chinese University of Hong Kong (CUHK), Prof. Chen joined CityU in 2012. Courses which he taught include Operations Management, Supply Chain Management, Logistics, and Advanced Manufacturing Management. He was also actively involved in executive teaching (EDP and EMBA). Prof. Chen has also been involved in consulting projects in the area of supply chain management and logistics. His current research projects span from healthcare operations management, logistics-supply chain management, to data-driven operations. He was project coordinators of two major projects which completed recently and has been principle investigator of more than 10 earmarked research grants.  
       

【講座信息】

  許多零售商會(huì)定期推出短生命周期的新產(chǎn)品,。不同于現(xiàn)有產(chǎn)品能夠根據(jù)歷史銷(xiāo)售數(shù)據(jù)來(lái)預(yù)測(cè)未來(lái)銷(xiāo)售，新產(chǎn)品沒(méi)有這樣的數(shù)據(jù),。取而代之的是,，一家公司過(guò)去可能一直在銷(xiāo)售類(lèi)似的產(chǎn)品，并很好地保存了銷(xiāo)售數(shù)據(jù),。除了需求/銷(xiāo)售數(shù)據(jù)外,，數(shù)據(jù)記錄還可能包含有關(guān)產(chǎn)品屬性（特征）的豐富信息，如零售價(jià)格,、設(shè)計(jì)風(fēng)格和季節(jié),，即所謂的需求協(xié)變量信息。在本研究中,，我們?cè)噲D通過(guò)使用協(xié)變量信息將一個(gè)新產(chǎn)品與歷史上銷(xiāo)售的“類(lèi)似”產(chǎn)品聯(lián)系起來(lái),。采用權(quán)重來(lái)度量新產(chǎn)品和歷史產(chǎn)品之間的相似性，將機(jī)器學(xué)習(xí)方法（如k近鄰法,、分類(lèi)回歸樹(shù)法和隨機(jī)森林法）應(yīng)用到數(shù)據(jù)中來(lái)估計(jì)權(quán)重值,。類(lèi)似歷史產(chǎn)品的現(xiàn)實(shí)需求及其對(duì)應(yīng)的權(quán)重，連同來(lái)自其他類(lèi)似產(chǎn)品的需求,，被用來(lái)近似估計(jì)期望利潤(rùn)和其他（按條件）需求分布的數(shù)量,。該方法應(yīng)用于風(fēng)險(xiǎn)規(guī)避企業(yè)在推出新產(chǎn)品前確定最優(yōu)訂貨量。風(fēng)險(xiǎn)規(guī)避要求企業(yè)獲得一個(gè)高置信度的利潤(rùn)目標(biāo),，該目標(biāo)可以表述為風(fēng)險(xiǎn)價(jià)值約束,。除了設(shè)計(jì)有效的解決方案外，我們還證明了所提出的近似估計(jì)方法是漸近最優(yōu)的,，即使是使用依賴(lài)于風(fēng)險(xiǎn)價(jià)值約束的樣本,。我們還將使用實(shí)際中的數(shù)據(jù)來(lái)驗(yàn)證我們的模型和方法,，并提出關(guān)鍵的管理啟示,。

Many retailers regularly introduce new, short life-cycle products. Unlike existing products whose historical sales data may be an indicator of future sales, a new product does not have such data. Instead, a firm may have been selling similar products in the past and keeps a good record of them. In addition to demand/sales figures, the data record may contain rich information about the attributes (features) of the products, such as retail price, design style, and season, the so-called covariate information to demand. In this project we attempt to link a new product, by using covariate information, to “similar” products that were sold historically. Weights are used to measure similarities between the new product and historical products, and the values of those weights are estimated by employing machine learning methods  such as   k-nearest neighbours, classification and regression tree, and random forests, to the data.  Then, the pair of the realized demand of a similar historical product and its associated weight, together with those from other similar products, are utilised to approximate the expected profit and other quantities which take on the (conditional) demand distribution. This approach is applied to determine the optimal order quantities before a risk-averse firm launches a new product. Risk aversion requires the firm to attain a profit target with high confidence, which can be formulated as a value-at-risk (VaR) constraint. Besides devising efficient solutions, we also prove the proposed approximation to be asymptotically optimal even with the sample-dependent approximation for the VaR constraint. We will also use real-world data to verify our models and methods and present key managerial insights.