ORC IAP Seminar 2024

1/30/24 | 10:00am-4:45pm | 45-432

OR Through the Ages

Description: This seminar will exhibit different focuses in operations research over time. We plan to have four talks throughout the day, each showcasing a distinct aspect of the field, as well as a panel discussion on the theme. 

Date: Tuesday, January 30th

Place: 45-432

Schedule 

 

Coffee Time - 9:30am-10:00am

 

Talk 1 - 10:00am-10:45am

Joey Huchette 

Software Engineer, Large Scale Optimization Team

Google Research

Title
Optimizing Compute Infrastructure at Scale
Abstract

Through a series of vignettes, we will showcase how techniques from operations research and mathematical optimization can be used to make computation more efficient and reliable, at scale. We will also touch on the practical considerations of using optimization in production environments, including engineering best practices for robust and scalable implementations, as well as the (sometimes unexpected) expectations and requirements of end-users.
Bio

Joey Huchette is a Software Engineer in the Large Scale Optimization Team at Google Research. Previously, he was an Assistant Professor in the Computational and Applied Mathematics Department at Rice University, and before that a postdoc with the Operations Research Team at Google Research. He received his PhD from the Operations Research Center at MIT. His work is primarily in the areas of integer programming and computational optimization, and has been recognized by the Beale — Orchard-Hays Prize, the INFORMS Computing Society Prize, and the COIN-OR Cup.
 

Talk 2 - 11:00am-11:45am

Konstantina Mellou

Senior Researcher

Microsoft Research Redmond

Title
Optimizing the Cloud Supply Chain
Abstract

In this talk, we give an overview of how Operations Research has enabled one of the biggest cloud providers in the world, Microsoft Azure, to achieve and sustain cost-effective operations. We tackle issues such as handling supply chain dependencies and multi-dimensional resource management under uncertainty, and demonstrate how we were able to minimize stranded power and other operational costs, leading to hundreds of million dollars savings. We conclude with recent work on utilizing Large Language Models (LLMs) towards improving explainability of our optimization solutions for datacenter operators.
Bio

Konstantina Mellou is a Senior Researcher in the Cloud Operations Research (CORE) group at Microsoft Research Redmond. She received her Ph.D. in Operations Research from MIT, and before that her degree in Electrical and Computer Engineering from the National Technical University of Athens in Greece. Her research interests are in the area of optimization and data-driven decision making with applications in cloud computing, transportation, and logistics. At Microsoft Research, she has been working on designing large-scale optimization solutions for cloud systems and applications; part of this work has been deployed in Microsoft Azure and Dynamics 365. 

 

Lunch Break - 12:00pm-1:30pm

Lunch will be provide to those that RSVP'd
In Lounge 45-417

 

Panel Discussion - 1:30pm-3:00pm

​Topic: OR Through the Ages

Adam Elmachtoub

Joey Huchette

Konstantina Mellou

Wei Sun

Talk 3 - 3:00pm-3:45pm

Adam Elmachtoub

Associate Professor of Industrial Engineering and Operations Research 

Columbia University

Title
The Surprising Power of Static Pricing in Classic OR Models
Abstract

In this talk, we survey several of our recent results on using static pricing strategies as alternatives to dynamic pricing. We focus on three fundamental settings: Erlang loss system (https://arxiv.org/abs/2302.11723), M/M/1 queue (https://arxiv.org/abs/2305.09168), and stochastic inventory control (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4470538). We provide strong, non-asymptotic theoretical guarantees that hold under general conditions, and provide a novel analysis that utilizes the true optimal policy as a benchmark, rather than the classic benchmark based on a deterministic upper bound. These are joint works with Jiaqi Shi, Jacob Bergquist, Harsh Sheth, and Yeqing Zhou.
Bio

Adam Elmachtoub is an Associate Professor of Industrial Engineering and Operations Research at Columbia University, where he is also a member of the Data Science Institute. He is also an Amazon Visiting Academic. His research spans two major themes: (i) designing machine learning and personalization methods to make informed decisions in industries such as retail, logistics, and travel (ii) new models and algorithms for revenue and supply chain management in modern e-commerce and service systems. He received his B.S. degree from Cornell and his Ph.D. from MIT ORC, both in operations research. He spent one year as a postdoc at the IBM T.J. Watson Research Center working in the area of Smarter Commerce. He has received an NSF CAREER Award, IBM Faculty Award, 1st place in the INFORMS JFIG (Junior Faculty) Paper Competition, Great Teacher Award from the Society of Columbia Graduates, and was on Forbes 30 under 30 in science.

 

Talk 4 - 4:00pm-4:45pm

Wei Sun

Senior Research Scientist

IBM T. J. Watson Research Center

Title

Counterfactual-driven Policy Learning

Abstract

With the abundance of available data, many enterprises seek to implement data-driven prescriptive analytics to help them make better decisions. Despite machine learning (ML) making huge strides in recent years, obstacles are still standing in the way of the widespread adoption of prescriptive analytics.  In this talk, I will introduce some work at IBM Research around counterfactual-driven policy learning that aims to address these challenges faced by enterprises. One framework is called the counterfactual prescriptive tree, which learns interpretable optimal policy from observational data. The proposed framework consists of a causal teacher model which produces counterfactual outcomes corresponding to different treatment actions, and a prescriptive student model which distills a set of optimized policies in the form of a tree. The prescriptive tree can be learned greedily for swift deployment. The optimal prescriptive tree can also be constructed by solving a scalable mixed-integer problem via column generation. I will highlight the results from a recent online test that shows a 7% increase in revenue over the legacy pricing benchmark, where we applied this solution to a large US airline in premium seat upsell. Lastly, I will showcase an LLM-based agent that allows non-ML business end users to interact with advanced AI models via a natural language interface, democratizing prescriptive analytics for strategic decision-making in enterprises. 
Bio

Wei is a Senior Research Scientist under AI Models at IBM T. J. Watson Research Center in Yorktown Heights, NY. Her research centers on the intersections of machine learning and optimization, including data-driven policy learning, constrained prediction, counterfactual inference, and reinforcement learning. Her work has been applied to solve real-world challenges of many companies in digital marketing, travel/transport, and financial services. Wei has published in top-tier ML and optimization venues including ICML, AAAI, and Management Science, and holds more than 20 patents. 

Wei graduated with a Ph.D. in Operations Research and an M.S. in Computational Design and Optimization from the Massachusetts Institute of Technology (MIT). She also has an M.S. in Computational Engineering and a B.Eng. in Electrical and Computer Engineering with First-Class Honors from the National University of Singapore (NUS).

 

If you have any questions, please contact us via email: orc_iapcoordinators@mit.edu.

 

Event Time: 

2023 - 09:30