Antonio Alcantara Mata: A Quantile Neural Network Framework for Two-stage Stochastic Optimization

 Two-stage stochastic programming is widely used for optimization under uncertainty, where decision-making is split into first-stage decisions and second-stage recourse actions adjusted after uncertainties are realized. Traditionally, these problems are solved using Sample Average Approximation (SAA), which models uncertainty via scenarios, leading to large and complex problems. To address this, surrogate models have been used to approximate the expected second-stage objective and guide first-stage decisions. In this seminar, we introduce a novel approach that leverages a quantile neural network to model the distribution of the second-stage objective. This enables optimization not only for expected values but also for risk-sensitive measures, such as Conditional Value at Risk (CVaR). We will discuss optimization formulations for embedding this distributional approximation and present computational results demonstrating the effectiveness of the approach on mixed-integer optimization problems.