ScholarWorks@숙명여자대학교

초록

Quantum neural networks (QNNs) have attracted growing interest for their potential to accelerate computation and leverage quantum supremacy. Their performance largely depends on gate placement within parameterized quantum circuits (PQCs), making neural architecture search (NAS) a suitable approach for discovering efficient structures. However, applying conventional NAS to QNNs is hindered by barren plateaus, noise, hardware constraints, and high computational costs. To address these challenges, this paper proposes filtered one-Shot training for quantum architecture search, which combines deep reinforcement learning (DRL) for constraint-aware gate placement with a one-shot supernet for efficient weight sharing. A filtering mechanism further removes weak paths to narrow the search space. Experimental results show that the method reduces parameters by up to 76% while maintaining high accuracy. © 2025 Copyright held by the owner/author(s).

키워드