Coupling-Grouped XY-QAOA for Joint Anomaly-Feature Selection

Abstract:Selecting anomalous samples and explanatory features under fixed cardinalities defines a coupled optimization problem: feature-first selection can overlook features whose usefulness depends on the selected samples. In the analyzed model, joint calibration-error sensitivity is sample-count independent, while the sum-column feature-first rule's sensitivity grows linearly. To optimize this formulation, we introduce Coupling-Grouped XY Quantum Approximate Optimization Algorithm (CG-XY-QAOA), with constraint-preserving mixers and block-structured phase angles. On matched sparse IBM Heron R3 circuits, our implementation reduces circuit depth by 35.1%-54.8% against Qiskit's optimization-level 3 preset pass manager on the CZ-basis target and enables runs at 64 decision qubits (p=2) and 36 (p=3). At 20 decision qubits (p=5), fully grouped angles raise the fixed low-energy hit probability 25-fold to 3.7%, with 64.7% of measured samples satisfying both constraints. Benchmarks favor joint selection when feature usefulness varies across anomalies. Noiseless simulations show problem-structured grouping improves over same-depth XY-QAOA and parameter-matched type-preserving randomizations.

arXiv:2606.13244 [quant-ph]

(or arXiv:2606.13244v2 [quant-ph] for this version)

https://doi.org/10.48550/arXiv.2606.13244 Focus to learn more arXiv-issued DOI via DataCite

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.