Quantum Neural Network Extraction Attack via Split Co-Teaching

Quantum Neural Networks (QNNs), now offered as QNN-as-a-Service (QNNaaS), have become key targets for model extraction attacks. State-of-the-art methods use ensemble learning to train accurate substitute QNNs, but our analysis reveals significant limitations in real-world NISQ environments, where noise and cost constraints undermine their effectiveness. In this work, we introduce a novel attack, split co-teaching, which leverages label variations to split queried data by noise sensitivity and employs co-teaching schemes to enhance extraction accuracy. Experiments on NISQ processors demonstrate that our approach outperforms classical extraction attacks by 6.5%~9.5% and existing QNN extraction methods by 0.1%~3.7% across various tasks.