Haihao Lu

Haihao Lu, incoming Assistant Professor of Operations Research and Statistics, won the 2024 Beale — Orchard-Hays Prize for Excellence in Computational Mathematical Programming from the Mathematical Optimization Society. One of the highest honors in the field of computational optimization, this prize is awarded every three years for a publication during the prior six years.

Lu’s winning papers are:

1) Faster First-Order Primal-Dual Methods for Linear Programming Using Restarts and Sharpness, co-authored with David Applegate (Google Research), Oliver Hinder (University of Pittsburgh), and Miles Lubin (Hudson River Trading; formerly Google Research); Mathematical Programming, Vol. 201, 133-184 (2023)

2) Practical Large-Scale Linear Programming Using Primal-Dual Hybrid Gradient, co-authored with David Applegate, Mateo Díaz (Johns Hopkins University), Oliver Hinder, Miles Lubin, Brendan O’Donoghue (Google DeepMind), and Warren Schudy (Google Research); Proceedings of the 35th Annual Conference on Neural Information Processing Systems (NeurIPS 2021)

The award recognizes “long-term potential to make first-order methods a practical option to solve large-scale linear programming … adaptability to GPUs and other parallel computing architectures; the careful algorithmic engineering work to make the methods practical; and the sophisticated and innovative analysis used to justify and describe the performance of the algorithms.”

Haihao Lu was awarded the INFORMS Revenue Management and Pricing Section Prize while he was an assistant professor at the University of Chicago Booth School of Business. Lu won the award jointly with Santiago Balseiro (Columbia University) and Vahab Mirrokni (Google Research). Their winning paper was entitled, “The Best of Many Worlds: Dual Mirror Descent for Online Allocation Problems.”

The citation of the prize said, “The paper unifies a diverse array of resource allocation problems from both modeling and algorithmic viewpoints. Looking at network revenue management, budgeted repeated auctions, online matching and assortment personalization problems through a single model, the authors use the celebrated bid price machinery to give provably optimal algorithms with matching lower and upper bounds on the performance loss. In doing so, the paper dramatically improves our technical understanding of bid price policies, while giving us tools highly relevant to the practice of revenue management.”

Haihao Lu was awarded the Michael H. Rothkopf Junior Researcher Paper Prize (first place) while he was an assistant professor at the University of Chicago Booth School of Business.

This prize recognizes the best paper by a young researcher in the area of Auctions and Market Design, with a particular focus on the use of Operations Research and/or Management Science techniques in the design or analysis of markets. The winning paper is entitled “The Best of Many Worlds: Dual Mirror Descent for Online Allocation Problems”, jointly with Santiago Balseiro (Columbia University) and Vahab Mirrokni (Google Research).

Haihao Lu was honored with the INFORMS Optimization Society Young Researcher Prize while he was an assistant professor at the University of Chicago Booth School of Business.

This prize serves as an esteemed recognition of promising colleagues in the field of optimization who are at the beginning of their academic or industrial career. The winning material is a solo-authored paper, entitled “An O(s^r)-Resolution ODE Framework for Understanding Discrete-Time Algorithms and Applications to the Linear Convergence of Minimax Problems.” The citation of the prize reads, “There has been renewed interest in using ordinary differential equations (ODEs) to understand the dynamics of a discrete-time optimization algorithm (DTA), due to the ease of analysis of ODEs compared to that of a DTA. However, two fundamental open questions have limited the applications of the ODE approach in optimization algorithm analysis: 1) how to obtain a suitable ODE from an optimization method? 2) what is the connection between the convergence of the ODE and the convergence of the optimization method? Haihao Lu’s paper constructively answers the first question by proposing a hierarchy of O(s^r)-resolution ODEs, parameterized by the order r, that converges to the discrete-time optimization method as r goes to infinity, where s is the step-size of the algorithm. Furthermore, the paper addresses the second question by showing that with a properly chosen energy function in the analysis, the linear convergence of the ODE automatically guarantees the linear convergence of the corresponding optimization method. Finally, the paper answers fundamental questions about three central algorithms for minimax problems and their convergent/divergent behavior. In summary, the novel and unified framework developed in this paper for analyzing a broad class of optimization algorithms makes it highly deserving of the INFORMS Optimization Society Young Researcher Prize.”