Explore a 27-minute DS4DM Coffee Talk on distributionally robust risk-averse network interdiction problems. Delve into the effectiveness of randomization strategies for interdictors who are both risk- and ambiguity-averse. Learn about the introduction of a distributionally robust maximum flow network interdiction problem that minimizes the worst-case Conditional Value at Risk (CVaR) of maximum flow. Discover how the problem is reformulated as a bilinear optimization problem and solved using a spatial branch-and-bound algorithm. Gain insights into the development of a column-generation algorithm for identifying optimal support and its application in coordinate descent for upper bound determination. Examine numerical experiments that demonstrate the efficiency and convergence of the proposed algorithm, as well as the superior performance of randomized strategies compared to deterministic ones.
The Value of Randomized Strategies in Distributionally Robust Risk-Averse Network Interdiction Problems
GERAD Research Center via YouTube
Overview
Syllabus
Introduction
Motivation
Ambiguity
Risk aversion
Model
Support Function
Global Optimal Solution
Lower Bound
Context Relaxation
Convex Relaxation
Choosing Midpoints
Numerical Experiments
Taught by
GERAD Research Center