New Publication(Training-Free Entangler Selection for Quantum Neural Networks via Hilbert–Schmidt Geometry) in NeurIPS 2026

We are pleased to announce that our team’s new paper has been accepted to NeurIPS 2026 (Conference on Neural Information Processing Systems), one of the most prestigious venues in machine learning.

This study addresses how to choose where entangling gates should be placed in quantum machine learning circuits—a central design choice that is still typically made from fixed templates or through expensive search procedures requiring training or pairwise kernel evaluations. The paper proposes the HSD-Lipschitz principle, a training-free geometric criterion built on a simple intuition: a useful entangler should keep encoded states stable across the dataset while separating different classes. This is formalized through a two-sided bound on local light-cone subsystems, where dataset-level gradient variance is upper-bounded by reduced-state spread and the expected between-class gradient signal is lower-bounded by class-conditional reduced-state distance. The resulting algorithm, HSD-greedy, adaptively grows an entanglement structure using only single-state reduced moments estimable via classical shadows, avoiding the O(N²) pairwise kernel evaluations of kernel-target alignment proxies. Both bounds hold without violation across 110 topologies and 440 parameter rows, and HSD-greedy attains the strongest average AUC among training-free entangler selectors on synthetic and real-world benchmarks at roughly 60× lower selection cost than the strongest baseline in 13-qubit simulation. End-to-end execution on a 20-qubit IBM Eagle subchain substantially outperforms a fixed 19-CNOT Linear chain while using far fewer two-qubit gates.

“Training-Free Entangler Selection for Quantum Neural Networks via Hilbert–Schmidt Geometry” Woo Seob Sim, & Yu Rang Park (2026). NeurIPS 2026.