Tepper School Doctoral Student Earns INFORMS Recognition
A paper co-written by Ph.D. student Nam Ho-Nguyen received an honorable mention for the 2018 INFORMS Optimization Society Student Paper Prize.
A paper co-written by Ph.D. student Nam Ho-Nguyen and his adviser, Fatma Kılınç-Karzan, Associate Professor of Operations Research, “Online First-Order Framework for Robust Convex Optimization,” was recognized with an honorable mention by the Optimization Society of the Institute for Operations Research and the Management Sciences. The INFORMS Optimization Society Student Paper Prize is an annual accolade for outstanding papers in optimization submitted or accepted for publication in a professional journal.
“The paper provides a new, unified framework for thinking about quantitative decision-making models when data are corrupted by noise,” Ho-Nguyen said. Noise refers to any errors that interfere with the accuracy of data, such as future estimates for the demand of a product or the ones due to adversarial nature of competition.
As data becomes integrated into the decision-making process in more and diverse fields, noise remains a significant challenge for analysts. While robust optimization has been useful in handling uncertainties like those caused by noise, the solutions tend to be computationally demanding, which is impractical for machine learning and statistics. “This leaves a particular and important question open,” Kılınç-Karzan said. “Can the overall computational effort for solving these problems be reduced to the point of successfully handing the large-scale applications arising in these fields?”
“By analyzing our framework more carefully we can propose new methods which eliminate some shortcomings of existing ones,” Ho-Nguyen said. “In particular, our methods work well for problems with a large number of decision variables, by utilizing popular first-order algorithms from large-scale optimization.”
Ho-Nguyen noted that the increased focus on big data across business functions has made their work more relevant to the field of optimization in general. “We hope that this leads to more people working on robust optimization using our framework, which may provide further exciting insights,” he said.
“This paper represents an impressive body of work written in a very eloquent way,” Kılınç-Karzan said. “Quite remarkably, in many cases, the algorithms resulting from this framework are almost as efficient as if there is no uncertainty involved. We are delighted to see the INFORMS Optimization Society acknowledge the significance of this work via this recognition."
Ho-Nguyen plans to continue studying models and computationally efficient solution techniques geared toward decision-making in the face of incomplete knowledge and/or dynamic environments with a focus on the problems arising from business analytics and machine learning. One particular area of interest is how to accurately and efficiently learn customer preferences from large-scale dynamic observational data, which is a problem faced by many retailers like Amazon.