Advising

I have been fortunate to work with a number of exceptional students. I greatly enjoy working with students and view this work as a highly collaborative process that allows us both to solve problems on the cutting edge of research. This page describes student research that I have advised.

Current Ph.D. students

  • Ye Chen (expected graduation: 2018). Ye is developing a new consistency theory for statistical estimators derived using the methodology of approximate Bayesian inference. This is a challenging open problem in the literature and has direct implications for learning models previously studied in revenue management, e-commerce, resource allocation and other applications.

  • Yimei Fan (expected graduation: 2017). Yimei is working on high-dimensional model selection in the context of massive marketing datasets where the goal is to model customer response to a large variety of products. In practice, the common characteristics of these products are often modeled using a hierarchy of binary variables; the challenge is to screen out as many of these variables as possible while retaining strong predictive power.

  • Liyi Gu (expected graduation: 2019). Liyi is working on both methodological and empirical topics in operations management. His work deals with applications in e-commerce as well as humanitarian logistics. Currently, we are developing a simulation model for evaluating assignment policies for vehicles in humanitarian delegations around the world.

  • Jinhang Xue (expected graduation: 2018). Jinhang is working on modeling uncertainty about convex functions, a general statistical problem that nonetheless has very direct applications in transportation and revenue management, where operating costs in a given region are often convex in the number of vehicles assigned.

Past Ph.D. students and their placement

Bin Han (2015). "Statistical and optimal learning with applications in business analytics." First position: Blackrock, Inc.

Bin's dissertation focused on both statistical and optimal learning, mainly in the context of non-profit management. We first worked on an empirical project in which we applied model selection methods to a massive dataset covering over 8 million recorded direct-mail interactions with donors to the American Red Cross (ARC). The results of this study, which were presented to ARC marketing executives and later published in Management Science, identified fundraising design practices that exerted a positive impact on response rates. In his second paper, Bin explored the closely related problem of optimizing future fundraisers based on the regression model trained from the data.

Bin won numerous awards for his work. Within the University of Maryland, he won the prestigious Ann G. Wylie Dissertation Fellowship (a full-semester scholarship), as well as the Phi Delta Gamma Graduate Fellowship. He was also recognized as a finalist in the 2015 INFORMS Washington DC Chapter Student Excellence Award competition.

  • Han, B., Ryzhov, I.O. & Defourny, B. (2016) "Optimal learning in linear regression with combinatorial feature selection." INFORMS Journal on Computing 28(4), 721-735. PDF
  • Ryzhov, I.O., Han, B. & Bradić, J. (2016) "Cultivating disaster donors using data analytics." Management Science 62(3), 849-866. PDF

Zi Ding (2014). "Optimal learning with non-Gaussian rewards." First position: Citadel, LLC

Zi studied the theoretical problem of information collection in a setting where the observed rewards are non-Gaussian. In the Gaussian case, one approach for calculating the optimal "Gittins index" policy is to construct a continuous-time Brownian motion whose increments have the same distribution as the rewards in the original discrete-time problem. Zi found that, for some non-Gaussian reward distributions, it is possible to conduct a similar analysis using a more general Lévy process model. His work provides the first theoretical characterization of the optimal policy in this continuous-time setting.

Zi's dissertation paper placed as a finalist in the 2014 INFORMS Junior Faculty Forum paper competition.

  • Ding, Z. & Ryzhov, I.O. "Optimal learning with non-Gaussian rewards." Advances in Applied Probability 48(1), 112-136. PDF

Huashuai Qu (2014). "Simulation optimization: new approaches and an application." First position: Google, Inc.

Huashuai made several contributions to simulation optimization methodology (some of them with Prof. Michael Fu). He worked with me on the ranking and selection problem, a fundamental model in the simulation literature for the study of information collection. When we are trying to learn about a large number of competing design alternatives (e.g., configurations for a simulator), and the available simulation budget is small, it is necessary to identify and exploit similarities and differences between alternatives in order to learn quickly; that is, a single simulation experiment involving one alternative should provide information about other, similar alternatives. Huashuai used approximate Bayesian inference to develop the first tractable statistical model for this problem, and showed how it can interface with efficient algorithms for allocating the simulation budget.

Huashuai's work on this topic won the 2012 INFORMS Computing Society Student Paper Award, as well as the Best Theoretical Paper Award at the 2012 Winter Simulation Conference.

  • Qu, H., Ryzhov, I.O., Fu, M.C. & Ding, Z. (2015) "Sequential selection with unknown correlation structures." Operations Research 63(4), 931-948. PDF