澳门2138cn太阳娱乐网站官网-在线登录

The growing availability of high-dimensional data has posed significant challenges to online learning and decision-making problems in data science. In this talk, we will address big data, modern statistical learning, and online decision making.

In the first part, we study a modification to the traditional sample average approximation (SAA) in the scenario where the global minimizer is either sparse or can be approximated by a sparse solution. We propose a regularization penalty referred to as the folded concave penalty (FCP), and show that, if an FCP-SAA is solved locally, then the required number of samples can be significantly reduced in approximating the global solution of a convex stochastic programming: the sample size is only required to be poly-logarithmic in the number of dimensions.

In the second part, we propose a minimax concave penalized multi-armed bandit algorithm under a generalized linear model, the G-MCP-Bandit algorithm, and demonstrate that the algorithm asymptotically achieves the optimal cumulative regret, with logarithmic dependence on both the sample size dimension and the covariate dimension. Through experiments based on synthetic data and real-world dataset, we show that the algorithm outperforms other benchmark algorithms in terms of cumulative regret and that the benefits of the G-MCP-Bandit algorithm seem to increase with the data's sparsity level and the size of the decision set. Finally, during a collaboration with one big E-commerce company, we adopted the algorithm to improve its C2C marketplace’s personalized recommendation system and observe that under high-dimensional settings (with millions of covariates and arms), the G-MCP-Bandit algorithm has the potential to substantially improve the platform’s prediction accuracy and revenue performance.

报告人简历：

Tao Yao is Principal Engineer and Chief Scientist of Operations Research at Alibaba Damo Academy. He has been

Associate Professor with Tenure at the Marcus Department of Industrial and Manufacturing Engineering at The

Pennsylvania State University. He holds a Ph.D. in Management Science and Engineering and a M.S. in Engineering-Economics System and Operations Research from Stanford University, a M.S. in Mathematics from UCLA, and a B.S. in Mathematics from Peking University.

Dr. Yao's areas of expertise include optimization, machine learning, data science, stochastic modeling, business analytics, and game theory. He has more than 60 articles published on refereed research journals and conference proceedings including Mathematical Programming, Annual of Statistics, Production and Operations Management, International Conference of Machine Learning, Transportation Research Part B: Methodological, SIAM Journal of

Control and Optimization, IEEE Transactions on Engineering Management. He has received research funding from NSF (5 regular grants), NSFC (1 overseas distinguished young scholar grant), MAUTC/USDOT (5 grants), and industry. He has received an honorable mention in the INFORMS George B. Dantzig Dissertation Award, several Best Paper Awards from IERC Supply Chain and Logistics track and CIS track, Best Paper Award (Finalist) from INFORMS Service Science section, and Best Student Paper Award (Runner-up) from INFORMS Conference on Information Systems and Technology.

He has been the Senior Editor for Production and Operations Management, on the Editorial Advisory Board of Transportation Research Part B: Methodological, the Area Editor for Network and Spatial Economics, and on the committee on Transportation Network Modeling, ADB30, Transportation Research Board of the National Academies. He has been very active with refereeing journal articles and grant proposals. He has served as the track chair for the Manufacturing Operations track of Production and Operations Management society (POMS) 2017 annual conference, the president of the Computer and Information Systems (CIS) Division of Institute of Industrial Engineering (IIE) and the chair of the 2011 Industrial Engineering Research Conference (IERC) CIS track.

Dr. Yao has led the Alibaba Decision Intelligence team of 100 algorithm engineers, focusing on AI research, product, and solution on optimization, machine learning, and data science to tackle complex real-world problems, including data intelligence, digital transformation, deep learning, intelligent manufacturing, supply chain and operations management, transportation and logistics, search and recommendation system, and automated reasoning.

欢迎您参与交流与讨论