Neural Heuristics for Mathematical Optimization via Value Function Approximation

Explore the cutting-edge application of neural heuristics in mathematical optimization through value function approximation in this 51-minute DS4DM Coffee Talk. Delve into Justin Dumouchelle's research from the University of Toronto, presented at the GERAD Research Center. Learn how a single learning-based framework can be adapted to stochastic optimization, bilevel optimization, and robust optimization, addressing real-world problems involving uncertainty and agent reactions. Discover the empirical findings that demonstrate solutions of similar or superior quality to state-of-the-art algorithms, often achieved in significantly less time. Gain insights into the open-sourced datasets and frameworks, including Neur2SP, Neur2RO, and Neur2BiLO, and their potential applications in various optimization domains.