Quantum-based self-attention mechanism for hardware-aware differentiable quantum architecture search
Phys. Rev. Applied 25, 064053 – Published 16 June, 2026
DOI: https://doi.org/10.1103/qnld-73pk
Abstract
The automated design of parameterized quantum circuits for variational algorithms in the noisy intermediate-scale quantum (NISQ) era faces a fundamental limitation, as conventional differentiable architecture searches rely on classical models that fail to represent quantum gate interactions adequately under hardware noise. We introduce Quantum-Based Self-Attention for Differentiable Quantum Architecture Search (QBSA-DQAS), a metalearning framework featuring quantum-based self-attention and hardware-aware multiobjective searching for automated architecture discovery. The framework employs a two-stage quantum-based self-attention module. First, it computes contextual dependencies by mapping architectural parameters through parameterized quantum circuits, extracting high-dimensional feature representations that replace classical similarity metrics with quantum-derived attention scores. Second, it applies positionwise quantum transformations for feature enrichment. Architecture searching is guided by a task-agnostic multiobjective function, jointly optimizing noisy expressibility and the probability of successful trials. A postsearch optimization stage applies gate commutation, fusion, and elimination to reduce circuit complexity. Experimental validation demonstrates superior performance in variational quantum eigensolver (VQE) tasks and large-scale wireless sensor networks (WSNs). For VQE tasks on , QBSA-DQAS achieves 0.9 accuracy, compared with 0.89 for a standard differentiable quantum architecture search, outperforming classical applied attention baselines. Postsearch optimization via deterministic simplification rules reduces the complexity of the discovered circuits by up to 44% in gate count and 47% in depth without accuracy degradation. The framework maintains robust performance across three molecules and five IBM Quantum hardware noise models. For WSN routing, discovered circuits achieve 8.6% energy reduction versus the quantum approximate optimization algorithm and 40.7% versus the classical greedy method. These results establish the effectiveness of quantum-native architecture searching for NISQ applications.
