[论文] Input-Aware Dynamic Backdoor Attack Against Quantum Neural Networ...

论文概要

研究领域: quantum 作者: Junrui Zhang, Zemin Chen, Lusi Li, Mohammad Ghasemigol, Daniel Takabi 发布时间: 2026-07-13 arXiv: 2607.11843

中文摘要

量子神经网络（QNN）是近-term量子设备上量子机器学习的有前景框架，但其安全风险仍未被充分理解。研究表明QNN易受后门攻击，但现有量子后门大多依赖所有中毒输入共享的固定触发器。这种固定触发器设计是主要弱点，因为许多防御措施检测或削弱触发器在数据表征中留下的重复模式。尽管输入感知动态后门在经典神经网络中已有研究，但迁移到QNN存在困难，因为量子学习引入了新的障碍：测量将后ansatz量子态压缩为有限经典输出，削弱触发器生成器的监督；而个体密度矩阵随输入波动，使每样本对比学习不稳定。本文提出Q-DIBA，首个面向QNN的输入感知动态后门攻击。Q-DIBA通过三模式小批量策略联合训练经典触发器生成器和受害QNN，支持干净行为、攻击激活和触发器特异性。为提供稳定的量子级监督，引入集成密度对比损失，在测量前作用于后ansatz量子态，对比模式平均密度矩阵而非个体样本。在MNIST和Fashion-MNIST跨多个QNN架构上的实验表明Q-DIBA实现高干净准确率、强攻击成功率和高跨触发器准确率，对视觉检查、谱签名检测和微调等防御仍具韧性。

原文摘要

Quantum Neural Networks (QNNs) are a promising framework for quantum machine learning on near-term quantum devices, but their security risks remain insufficiently understood. Studies have shown that QNNs are vulnerable to backdoor attacks, yet existing quantum backdoors mostly rely on a fixed trigger shared by all poisoned inputs. This fixed-trigger design is a major weakness because many defenses detect or weaken the repeated patterns such triggers leave in data representations. Although input-aware dynamic backdoors have been studied in classical neural networks, transferring them to QNNs is difficult because quantum learning introduces new obstacles. In particular, measurement compresses the post-ansatz quantum state into a limited classical output, weakening supervision for a trigger g...

--- *自动采集于 2026-07-15*

#论文 #arXiv #quantum #小凯