Scholarlink

<p>As deep learning models grow in complexity, they suffer from increasing redundant parameters and high computational costs. Quantum computing offers a new computational paradigm; however, fully quantum AI remains impractical due to current quantum hardware constraints. This raises a fundamental problem: Can quantum AI serve as an efficient module within classical AI architectures to enhance performance while reducing computational overhead? In this work, we explore a hybrid quantum-classical AI framework that embeds parameter-efficient quantum computing modules within classical deep learning models. Specifically, we incorporate a quantum bottleneck module into a U-Net architecture, called QB-Net. Experiments and quantitative analysis in real-world datasets for medical and natural image processing demonstrate that QB-Net reduces up to 36.3<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> parameters in the bottleneck versus its classical counterpart, while achieving performance comparable to classical U-Net in all metrics considered. Our results demonstrate that quantum AI can serve as a plug-and-play enhancement, leveraging quantum advantages to optimize classical architectures without altering their fundamental structure.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

HiQC: hybrid quantum classical deep learning with parameter efficient quantum modules

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • Liuke Xu,
  • Taotao Zhao,
  • He Li

摘要

As deep learning models grow in complexity, they suffer from increasing redundant parameters and high computational costs. Quantum computing offers a new computational paradigm; however, fully quantum AI remains impractical due to current quantum hardware constraints. This raises a fundamental problem: Can quantum AI serve as an efficient module within classical AI architectures to enhance performance while reducing computational overhead? In this work, we explore a hybrid quantum-classical AI framework that embeds parameter-efficient quantum computing modules within classical deep learning models. Specifically, we incorporate a quantum bottleneck module into a U-Net architecture, called QB-Net. Experiments and quantitative analysis in real-world datasets for medical and natural image processing demonstrate that QB-Net reduces up to 36.3 \(\times \) × parameters in the bottleneck versus its classical counterpart, while achieving performance comparable to classical U-Net in all metrics considered. Our results demonstrate that quantum AI can serve as a plug-and-play enhancement, leveraging quantum advantages to optimize classical architectures without altering their fundamental structure.