Skip to main content

IEOR Seminar by Dr. Sriram Sankaranarayanan

Title: Bilevel programming representability and algorithms for Nash games among bilevel leaders

Venue: IEOR Teaching Lab (Room 011,  Ground Floor, IEOR Building)

Date & Time: November 13,  2019, 10 AM

Abstract: Games where sequential decision making occurs between two players with different objectives can be posed as a mathematical programming problem called the bilevel program. In the first part of the talk, we provide a characterization of the structure of the feasible set of a mixed-integer bilevel linear program, in its full generality as well as in interesting restricted cases. In the second part of the talk, we analyze a subset of equilibrium problems with equilibrium constraints (EPECs), which are a special form of bilevel games. We prove computational complexity lower bounds to decide existence of Nash equilibrium for these problems. Then using some of the results from the first part of the talk, we provide the first algorithm to identify a Nash equilibrium for these problems. We also show that the time complexity of these algorithms match the proven lower bounds (up to lower order terms) and provide computational test results for the algorithm and its enhancements. Finally, we use this structure to model energy trade between Chile and Argentina, and identify domestic policies and trade policies that can help the countries reduce their greenhouse gas emissions.

About the Speaker: Sriram Sankaranarayanan is a postdoctoral fellow at Polytechnique Montréal in the Canada Excellence Research Chair in Data Science for Real-time Decision making, working with Prof. Andrea  Lodi. Before that, he had completed his bachelors degree at IIT Kharagpur and his PhD at the Johns Hopkins University, guided by Prof. Sauleh Siddiqui and Prof. Amitabh Basu. His work is focused on developing and improving the algorithms for structured optimization and game theoretic problems, and applying them to large-scale real-life problems, particularly in the energy sector.

 

News Category
Date Posted