{"paper":{"title":"Learning Scenario Reduction for Two-Stage Robust Optimization with Discrete Uncertainty","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A GNN-Transformer trained to imitate a lookahead heuristic selects reduced scenarios for two-stage robust optimization while matching quality at 7-200x higher speed.","cross_cats":["cs.LG"],"primary_cat":"cs.AI","authors_text":"Jianan Zhou, Jieyi Bi, Jie Zhang, Tianjue Lin, Wen Song, Yaoxin Wu, Zhiguang Cao","submitted_at":"2026-05-14T07:34:13Z","abstract_excerpt":"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 yie"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"NeurPRISE consistently achieves competitive regret relative to comprehensive methods, maintains strong scalability 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.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the marginal impact of each scenario on the recourse cost can be accurately approximated by a GNN-Transformer trained only on PRISE's selections, without needing to solve the full subproblems at inference time, and that this approximation transfers across problem scales and distributions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"NeurPRISE trains a GNN-Transformer via imitation learning to mimic a lookahead heuristic for scenario reduction in 2RO, delivering 7-200x speedups with competitive regret on three test problems and zero-shot generalization.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A GNN-Transformer trained to imitate a lookahead heuristic selects reduced scenarios for two-stage robust optimization while matching quality at 7-200x higher speed.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"57f20ffff208781e1992a28627b8486606ca447d3768eddf4cbd52b165fc65e5"},"source":{"id":"2605.14494","kind":"arxiv","version":1},"verdict":{"id":"62f87a8e-2a1d-46b5-8c7f-f73965a6069c","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:00:16.509140Z","strongest_claim":"NeurPRISE consistently achieves competitive regret relative to comprehensive methods, maintains strong scalability 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.","one_line_summary":"NeurPRISE trains a GNN-Transformer via imitation learning to mimic a lookahead heuristic for scenario reduction in 2RO, delivering 7-200x speedups with competitive regret on three test problems and zero-shot generalization.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the marginal impact of each scenario on the recourse cost can be accurately approximated by a GNN-Transformer trained only on PRISE's selections, without needing to solve the full subproblems at inference time, and that this approximation transfers across problem scales and distributions.","pith_extraction_headline":"A GNN-Transformer trained to imitate a lookahead heuristic selects reduced scenarios for two-stage robust optimization while matching quality at 7-200x higher speed."},"references":{"count":37,"sample":[{"doi":"","year":2000,"title":"Robust solutions of linear programming problems contaminated with uncertain data.Mathematical programming, 88(3):411–424, 2000","work_id":"dd148461-db5f-48e9-a94a-5830912210c7","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2004,"title":"Adjustable robust solutions of uncertain linear programs.Mathematical programming, 99(2):351–376, 2004","work_id":"2fd54eec-a62e-4aa2-8c28-a355dd7bb338","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2009,"title":"Princeton University Press","work_id":"58a27603-78b1-4098-b51c-1b4ea23e8fa8","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2010,"title":"Finite adaptability in multistage linear optimization.IEEE Transactions on Automatic Control, 55(12):2751–2766, 2010","work_id":"24327f44-5d9f-4805-a4cb-40d9b227382f","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2015,"title":"Design of near optimal decision rules in multistage adaptive mixed-integer optimization.Operations Research, 63(3):610–627, 2015","work_id":"dd3882c4-c615-43f7-bf55-efc5178e92f2","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":37,"snapshot_sha256":"883ce7cab3cd6071d374c4d27beb2ac16c5122a84a7bd1643847ae00a2d20ec5","internal_anchors":1},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}