EURO 2025 Leeds
Abstract Submission

2534. Integration of Neural Network Surrogates in Two-Stage Stochastic Programming Problems

Invited abstract in session MB-31: Machine Learning for Optimization under uncertainty 1, stream Stochastic and Robust optimization.

Monday, 10:30-12:00
Room: Maurice Keyworth 1.06

Authors (first author is the speaker)

1. Enza Messina
DISCo - Department of Informatics, Systems and Communication, University of Milano Bicocca
2. Xiaochen Chou
Department of Informatics, Systems and Communication, University of Milano-Bicocca
3. Ludovica Di Marco
University of Milano Bicocca
4. Iman Seyedi
University of Milano Bicocca

Abstract

Decision-making under uncertainty addresses real-world problems that are often hard to solve. Stochastic programming is a practical approach in which decisions are determined prior to the realization of uncertain variables, with subsequent adjustments made through recourse mechanisms once the uncertainties are revealed. An optimal solution must ensure robustness and adaptability across a range of potential future scenarios. Traditional solving methods rely on scenario-based approximations, where incorporating additional scenarios may improve accuracy but simultaneously escalates computational complexity, often making large-scale problems infeasible.
With advancements in machine learning, this study aims to address these challenges by training neural networks as surrogate models for the recourse problems. These models estimate solution quality across diverse scenarios, thereby reducing the computational burden when integrated into the optimization framework. Empirical evaluations have been conducted on two-stage single-source capacitated Facility Location Problem, a stochastic variation of the multi-path Traveling Salesman Problem, and closed-loop Supply Chain Problem. Preliminary results demonstrate the effectiveness of this approach in balancing efficiency and accuracy taking advantage of the generalization capability of the neural network model. Future research will focus on extending its application to multi-stage stochastic programming problems.

Keywords

Status: accepted

Back to the list of papers