New DSI Member Vineet Goyal Designs Algorithms for Sequential Learning and Optimization
Vineet Goyal is one of the newest members of the Data Science Institute (DSI) at Columbia University and he possesses a multitude of talents that will enrich the institute’s education and research endeavors.
Goyal received his bachelor’s degree in computer science from the Indian Institute of Technology, Delhi, and his doctorate in algorithms, combinatorics, and optimization from Carnegie Mellon University. Before coming to Columbia, he was a postdoctoral associate at the Operations Research Center at MIT. Today, as an associate professor in Columbia’s Department of Industrial Engineering and Operations Research (IEOR), he focuses on designing algorithms for sequential learning and optimization under uncertainty, work that has wide applications in revenue management, resource allocation, health care, and energy markets. He has received the National Science Foundation (NSF) Career Award and his research has been supported by grants from the NSF, the Defense Advanced Research Projects Agency (DARPA), and industry partners, including Google, IBM, Adobe, and Amazon.
“Our society faces increasingly complex problems in a highly interconnected world, and interdisciplinary collaborations are crucial to working towards solving these problems," Goyal said. "In being the hub that all sectors turn to for data science expertise, DSI helps the university become significantly greater than the sum of its parts.”
His curiosity is wide ranging. For one research project, Goyal studies online advertising. It’s hard to assess the effectiveness of online advertising, he said, especially knowing what element of the campaign—the emails, display ads, or search ads—actually caused a consumer to buy a product. Several measures are used to attribute success to campaign elements, but there’s no current way to prove which element prompted the consumer to act. Goyal and his collaborators, DSI members Omar Besbes and Garud Iyengar, have developed a framework to allow advertisers to attribute success to the right elements of their advertising campaigns. Specifically, they use a Markovian model to track the consumer’s journey through the cycle of an advertising campaign to find proper attribution.
“Attribution is a fundamental question in many applications where a combination of several actions produces an end result and understanding how much each individual action influenced the result can have a significant impact in practice," Goyal said.
In another project, he studies a sequential resource allocation problem where the goal is to efficiently use a fixed capacity of resources to serve an uncertain demand. This is a classical problem arising in many applications. But this work with collaborators Iyengar and postdoctoral researcher Rajan Udwani is motivated by modern applications of sharing economies where resource capacities can be reused after they have finished serving a demand, for instance, servers in cloud computing. Several elements make this problem challenging, such as not knowing what resources will be needed to serve subsequent demand while making the allocation decisions for the current request and the service times of current demand. These resource requirements are often variable and difficult to predict at a fine granularity. To address the problem, Goyal and his collaborators designed an algorithm that doesn’t require any prediction of future job requests, yet can still compete with a benchmark that has full information about the future sequence of jobs.
“Not relying on predictions about the future is a very useful property in practice, especially in settings where the dynamics are very noisy and prediction is hard,” he said.
Goyal has joined five DSI research centers—Data, Media and Society, Financial and Business Analytics, Foundations of Data Science, Health Analytics, Smart Cities—and the Computational Social Science working group. He teaches core IEOR courses on the foundations of optimization, an analytics course on pricing and resource allocation problems that is an elective for M.S. in data science students, and a doctoral-level course that satisfies a requirement for the Ph.D. concentration in data science. He was also faculty adviser for a DSI capstone project in the fall of 2019 conducted in collaboration with Didi, a DSI affiliate.
"DSI increases the possibility of interdisciplinary research at Columbia by a significant degree of magnitude," he said while noting how delighted he is to be a member of the institute. "Algorithms and optimization are the foundational pillars of data science, and therefore my research aligns perfectly with the core mission of DSI.”
Media Contact: Vineet Goyal, firstname.lastname@example.org