ISyE Student Seminar

Under construction...

Sept 16th | Jie Wang | Sinkhorn Distributionally Robust Optimization

Abstract: We study distributionally robust optimization with Sinkorn distance - a variant of Wasserstein distance with entropic regularization. We derive convex programming dual reformulations when the nominal distribution is an empirical distribution and a general distribution, respectively. Compared with Wasserstein DRO, its reformulation is tractable for a larger class of loss functions, and the worst-case distribution is not necessarily finitely supported when the nominal distribution is so. Connections with the KL-divergence DRO and Wasserstein DRO models are also uncovered. We propose an efficient batch gradient descent with bisection search method for solving the dual reformulation. Finally, we provide various numerical examples using both synthetic and real data to demonstrate its competitive performance.



Past Talks

  • Mar. 29th: Alfredo Torrico. Greedy Adapts to Sharpness.

  • Mar. 8th: He Wang. How to prepare for IE/OR academic job market?

  • Mar. 1st: Emma Johnson. Pyomo tutorial.

  • Feb. 22nd: Ahmad Baubaid. Service Network Design for Less-than-Truckload (LTL) Carriers.

  • Feb. 15th: Paritosh Ramanan. An Asynchronous, Decentralized Solution Framework for the Large Scale Unit Commitment Problem.

  • Feb. 8th: Asteroide Santana. Challenges of Getting Started with Academic Research (part 2). Slides. DearFutureMe.

  • Feb. 1st: Thinh Doan. Distributed consensus-based methods over multi-agent systems.

  • Jan. 25th: Felipe Lagos. The Continuous Time Inventory Routing Problem.

  • Jan. 18th: Timur Tankayev. Models and Algorithms for Distributed Order Management.

  • Dec. 1st: Asteroide Santana, Alfredo Torrico. Challenges of Getting Started with Academic Research. Slides.

  • Nov. 16th: German Schnaidt. Non-Parametric Density Estimation for Randomly Censored Data Using Wavelets.

  • Nov. 9th: Jiaming Liang. A Doubly Accelerated Inexact Proximal Point Method .

  • Nov. 2nd: William Kong. Tutorial on First-Order Methods for Nonconvex Optimization .

  • Oct. 26th: Edward He. Dynamic Discretization Discovery Algorithms for Time-Dependent Shortest Path Problems.

Seminar History and Acknowledgements