{"paper":{"title":"NeuroRisk: Physics-Informed Neural Optimization for Risk-Aware Traffic Engineering","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"NeuroRisk embeds the Sort-and-Select structure of risk-aware traffic engineering into a neural unrolled optimizer to deliver solver accuracy at 100- to 100000-fold speedups.","cross_cats":["cs.LG"],"primary_cat":"cs.NI","authors_text":"Jiashuai Liu, Jingyi Cheng, Qiaozhu Zhai, Shizhen Zhao, Siyuan Feng, Ximeng Liu, Xiyuan Liu, Yike Liu, Yingming Mao, Yuzhou Zhou, Zhen Yao","submitted_at":"2026-05-13T01:17:45Z","abstract_excerpt":"In production Wide-Area Networks (WANs), correlated failures dominate availability losses, forcing operators to reserve large safety margins that leave substantial capacity underutilized. Achieving high utilization under strict availability targets therefore requires risk-aware Traffic Engineering (TE) over dozens to hundreds of probabilistic failure scenarios-yet solving this problem at operational timescales remains elusive. We demonstrate that existing risk-aware formulations can be unified under an embedded Sort-and-Select structure, exposing a fundamental trade-off between expressiveness "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"NeuroRisk achieves small optimality gaps relative to the solver with orders of magnitude speedup (10^2-10^5 ×) on risk objectives, while outperforming neural baselines on nominal throughput.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the Sort-and-Select structure can be faithfully embedded into a neural unrolled optimizer using gated edge-local reservations and permutation-invariant cues so that feasibility is enforced under explicit capacity constraints and scenario-dependent risk.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"NeuroRisk is a physics-informed deep unrolled optimizer for risk-aware traffic engineering that achieves small optimality gaps and 100-100000x speedup over solvers while outperforming neural baselines on throughput.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"NeuroRisk embeds the Sort-and-Select structure of risk-aware traffic engineering into a neural unrolled optimizer to deliver solver accuracy at 100- to 100000-fold speedups.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e86490f2147ee999e068c940310cf9fb1998631ef7cd6af1113b929dcc54523b"},"source":{"id":"2605.12862","kind":"arxiv","version":1},"verdict":{"id":"81cd1512-f332-47da-b88a-4e9a26af3cce","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T18:54:42.991424Z","strongest_claim":"NeuroRisk achieves small optimality gaps relative to the solver with orders of magnitude speedup (10^2-10^5 ×) on risk objectives, while outperforming neural baselines on nominal throughput.","one_line_summary":"NeuroRisk is a physics-informed deep unrolled optimizer for risk-aware traffic engineering that achieves small optimality gaps and 100-100000x speedup over solvers while outperforming neural baselines on throughput.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the Sort-and-Select structure can be faithfully embedded into a neural unrolled optimizer using gated edge-local reservations and permutation-invariant cues so that feasibility is enforced under explicit capacity constraints and scenario-dependent risk.","pith_extraction_headline":"NeuroRisk embeds the Sort-and-Select structure of risk-aware traffic engineering into a neural unrolled optimizer to deliver solver accuracy at 100- to 100000-fold speedups."},"references":{"count":54,"sample":[{"doi":"","year":2021,"title":"Firas Abuzaid, Srikanth Kandula, Behnaz Arzani, Ishai Menache, Matei Zaharia, and Peter Bailis. 2021. Contracting Wide-area Network Topologies to Solve Flow Problems Quickly. In18th USENIX Sym- posium","work_id":"24140772-4d81-4658-99cc-e5cdc8ba76e0","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Akyildiz, Ahyoung Lee, Pu Wang, Min Luo, and Wu Chou","work_id":"87a4dc94-06d8-41da-a9c4-b8c2e233532f","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1016/j.comnet","year":2014,"title":"A roadmap for traffic engineering in SDN-OpenFlow networks. Comput. Netw.71 (Oct. 2014), 1–30. https://doi.org/10.1016/j.comnet. 2014.06.002","work_id":"a069373b-7b31-499b-86df-c90e783d0fd2","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1145/2619239.2626316","year":2014,"title":"Conga: distributed congestion-aware load balancing for datacenters","work_id":"cf63b272-0106-4845-9120-5f9398da4fce","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"10.1145/3651890.3672237","year":2024,"title":"Rdma over ethernet for distributed training at meta scale","work_id":"7624824b-e6d7-482d-9b94-0e5d49db10f4","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":54,"snapshot_sha256":"a5d500aedef77bf32d2aaad11dc71a7a963c4308bd7130ce58dcf6a4810b55de","internal_anchors":2},"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"}