INFORMS Annual Meeting 2015
Operations research and analytics conference
Prof. Charles Nicholson and Weili Zhang will be attending the 2015 INFORMS Annual Meeting in Philadelphia November 1-4 . This conference is one of the top opportunities to connect with leading researchers and practitioners in the field of operations research (OR), management sciences (MS), and analytics. Approximately 5,000 attendees comprised of OR/MS faculty, students, and industry leaders will gather to present their work and connect with their peers.
Dr. Nicholson is the Optimization Network session chair and is presenting Optimal Flow Analysis a novel statistical learning perspective on a classical NP-hard OR problem. Weili Zhang is presenting a novel application based on this same research entitled Regression Based Relaxation.
Title: Optimal Flow Analysis, Prediction and Application
Presenting Author: Charles Nicholson, Assistant Professor, University of Oklahoma
Co-Author: Weili Zhang, Graduate Research Assistant, University of Oklahoma
Session: Optimization Network
Date: Wednesday Nov 04, 12:45 – 14:15
The fixed charge network flow problem is a classic NP-hard problem with many applications. To the best of our knowledge, this is the first paper that employs statistical learning technique to analyze the characteristics of optimal solutions. We develop an accurate propensity model based on this analysis to predict which arcs will have positive flow in an optimal solution. This propensity score allows for multiple applications such as identification of critical components in complex networks.
Title: Regression Based Relaxation Solution Approach For Fixed Charge Network Flow Problem
Presenting Author: Weili Zhang, Graduate Research Assistant, University of Oklahoma
Co-Author: Charles Nicholson, Assistant Professor, University of Oklahoma
Session: Optimization Heuristic Programming
Date: Wednesday Nov 04, 12:45 – 14:15
In this paper, a novel solution approach to the fixed charge network flow (FCNF) problem named regression based relaxation (RBR) is developed. RBR employs the probability of arc usage to form a new linear programming problem. Through rigorous testing, RBR outperforms linear programming relaxation and relaxation induced neighborhood search regardless of the complexity of the problem. In addition, the improvement of integrating RBR in the exact solver is robust for large FCNF problems.