{"paper":{"title":"Neural QAOA$^{2}$: Differentiable Joint Graph Partitioning and Parameter Initialization for Quantum Combinatorial Optimization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A neural generator learns graph partitions and QAOA starting parameters together by back-propagating through a differentiable quantum evaluator.","cross_cats":["cs.AI"],"primary_cat":"quant-ph","authors_text":"Jiahao Wu, Shengcai Liu, Zubin Zheng","submitted_at":"2026-05-13T06:43:10Z","abstract_excerpt":"The quantum approximate optimization algorithm (QAOA) holds promise for combinatorial optimization but is constrained by limited qubits. While divide-and-conquer frameworks like QAOA$^{2}$ address scalability by partitioning graphs into subgraphs, existing methods suffer from two fundamental limitations: i) misalignment between heuristic partitioning metrics and quantum optimization goals, and ii) topology-blind parameter initialization that leads to optimization cold starts. To bridge these gaps, we propose Neural QAOA$^{2}$, an end-to-end differentiable framework that jointly generates graph"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"our gradient-driven approach broadly outperforms heuristic baselines, ranking first on 101 instances. It exhibits zero-shot generalization across out-of-distribution graph topologies and scales.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The differentiable quantum evaluator acts as a high-fidelity performance surrogate that supplies accurate gradient guidance for the joint generator.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"A differentiable generative evaluative network jointly learns graph partitions and QAOA parameter initializations, outperforming heuristic baselines on 101 of 183 tested QUBO, Ising, and MaxCut instances with zero-shot generalization.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A neural generator learns graph partitions and QAOA starting parameters together by back-propagating through a differentiable quantum evaluator.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"070918e6bfe9c44be266967f6f9526e24190f10f404a0e5bb62efa21f0685ebf"},"source":{"id":"2605.13072","kind":"arxiv","version":1},"verdict":{"id":"5f25995d-9dd4-4ecc-a0ed-b454f6c98a41","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:07:48.107641Z","strongest_claim":"our gradient-driven approach broadly outperforms heuristic baselines, ranking first on 101 instances. It exhibits zero-shot generalization across out-of-distribution graph topologies and scales.","one_line_summary":"A differentiable generative evaluative network jointly learns graph partitions and QAOA parameter initializations, outperforming heuristic baselines on 101 of 183 tested QUBO, Ising, and MaxCut instances with zero-shot generalization.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The differentiable quantum evaluator acts as a high-fidelity performance surrogate that supplies accurate gradient guidance for the joint generator.","pith_extraction_headline":"A neural generator learns graph partitions and QAOA starting parameters together by back-propagating through a differentiable quantum evaluator."},"references":{"count":17,"sample":[{"doi":"","year":null,"title":"We collect all applicants for partitionjinto a candidate setU j","work_id":"d61c6981-f3bc-4f51-b3ae-10fb785aa32f","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"• If|U j| ≤C (j) remain, all candidates are accepted","work_id":"326d9711-342a-4561-8a4f-143781d3b266","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2013,"title":"weakly-coupled","work_id":"a24ebc77-03ce-4f15-aac5-0d3e611f5133","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"These features are standardized (zero mean, unit variance) within each graph instance to handle variations in graph scale","work_id":"991ce594-5d46-4cfd-ae9f-e3f510ff4cb5","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Edge weight normalization: Edge weights are normalized by the maximum absolute weight in the graph, ensuring wij ∈[−1,1]. Construction of the Offline Dataset (Doffline).To pre-train the quantum evalua","work_id":"fd5b4e8e-2caf-459f-bdd2-a5c869819e1f","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":17,"snapshot_sha256":"0b1edc318022d2de88369e80a3d946820751f2ef49d049b425c7e066f2712ec8","internal_anchors":0},"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"}