Kimiya Jozani | Searcharxiv

arXiv2026

Learning to Optimize at Scale: A Benders Decomposition-TransfORmers Framework for Stochastic Combinatorial Optimization

We propose a learning-augmented Benders decomposition framework to solve large-scale two-stage stochastic mixed-integer programs. We focus on the two-stage stochastic capacitated lot-sizing problem (TSSCLSP) under demand uncertainty. Our method accelerates the convergence of the decomposition by using a pre-trained TransfORmer model to rapidly generate high-quality approximate solutions for the scenario subproblems. This hybrid strategy uses the TransfORmer predictions to generate strong optimality and feasibility cuts, effectively guiding the Benders master problem. Our framework includes a novel expandable generation mechanism, allowing a model trained on a fixed horizon to solve instances of arbitrary length. For the test set considered, our method solves instances up to T = 270, a scale previously intractable for this approach, while maintaining zero infeasibility in the generated subproblem solutions. This demonstrates the potential of TransfORmers as powerful surrogate solvers embedded within classical decomposition algorithms.

math.OC↗

arXiv2025

A Multi-echelon Demand-driven Supply Chain Model for Proactive Optimal Control of Epidemics: Insights from a COVID-19 Study

Timely and effective decision-making is critical during epidemics to reduce preventable infections and deaths. This demands integrated models that jointly capture disease dynamics, vaccine distribution, regional disparities, and behavioral responses. However, most existing approaches decouple epidemic forecasting from logistics planning, hindering adaptive and regionally responsive interventions. We propose a novel epidemiological-optimization framework that jointly models epidemic progression and a multiscale vaccine supply chain. The model incorporates spatio-temporally varying effective infection rates to reflect regional policy and behavioral dynamics. It supports coordinated, data-driven decision-making across spatial scales through two formulations: a multi-objective Gini-based model and a knapsack-based model that leverages regional vulnerability indicators for tractability and improved mitigation. To address computational complexity, we design two scalable heuristic decomposition algorithms inspired by the Benders decomposition. The model is validated using COVID-19 data in the U.S.. We introduce SARIMA-based forecasting as a novel approach for validating epidemic-optimization models under data limitations. The results show that our approach can prevent more than 2 million infections and 30,000 deaths in just six months while significantly improving the accessibility of vaccines in underserved regions. Our framework demonstrates that integrating fairness and epidemic dynamics with vaccine logistics leads to superior outcomes compared to traditional myopic policies. Fairness improves overall efficiency in the long term by prioritizing the most vulnerable populations, leading to better long-term public health outcomes. The model offers policymakers a scalable and operationally relevant tool to strengthen preparedness and ensure a more effective and equitable response to epidemics.

cs.SI↗