Learning Scenario Reduction for Two-Stage Robust Optimization with Discrete Uncertainty
About
Two-Stage Robust Optimization (2RO) with discrete uncertainty is challenging, often rendering exact solutions prohibitive. Scenario reduction alleviates this issue by selecting a small, representative subset of scenarios to enable tractable computation. However, existing methods are largely problem-agnostic, operating solely on the uncertainty set without consulting the feasible region or recourse structure. In this paper, we introduce PRISE, a problem-driven sequential lookahead heuristic that constructs reduced scenario sets by evaluating the marginal impact of each scenario. While PRISE yields high-quality scenario subsets, each selection step requires solving multiple subproblems, making it computationally expensive at scale. To address this, we propose NeurPRISE, a neural surrogate model built on a GNN-Transformer backbone that encodes the per-scenario structure via graph convolution and captures cross-scenario interactions through attention. NeurPRISE is trained via imitation learning with a gain-aware ranking objective, which distills marginal gain information from PRISE into a learned scoring function for scenario ranking and selection. Extensive results on three 2RO problems show that NeurPRISE consistently achieves competitive regret relative to comprehensive methods, maintains strong calability with varying numbers of scenarios, and delivers 7-200x speedup over PRISE. NeurPRISE also exhibits strong zero-shot generalization, effectively handling instances with larger problem scales (up to 5x), more scenarios (up to 4x), and distribution shifts.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Scenario Reduction | VC Medium-scale 50 instances (test) | REG Score2.28 | 28 | |
| Scenario Reduction | VC Large-scale 50 instances (test) | Registration Error (REG)2.88 | 28 | |
| Scenario Reduction | SEL Medium-scale 50 instances (test) | REG Score6.64 | 28 | |
| Scenario Reduction | SEL Large-scale 50 instances (test) | REG5.8 | 28 | |
| Vertex Cover | VC-20-50 small (test) | REG1.04 | 26 | |
| Selection | SEL-20-50 small (test) | REG Score5.93 | 26 | |
| SEL | problem size instances SEL 5x scale (test) | Regret (%)0.81 | 9 | |
| VC | VC problem size instances 5x scale (test) | Regret (%)4.2 | 9 | |
| Capacitated Facility Location Problem | CFLP Small scale | Regret0.00e+0 | 5 | |
| Capacitated Facility Location Problem | CFLP Medium scale | Regret0.00e+0 | 5 |