Q-Detection: A Hybrid Quantum-Classical Poisoning Attack Detection Method

Total: 1

Data poisoning attacks pose a serious threat to machine learning models by introducing malicious data into the learning process, degrading model performance or manipulating predictions. Detecting and filtering out contaminated data is an important way to prevent data contamination attacks. Due to the limitations of existing computational frameworks, future large and complex datasets may pose a challenge for detection. For the first time, we present the unique speedup of quantum computing in data poisoning detection tasks. We present Q-Detection, a quantum-classical hybrid defense method for detecting poisoning attacks. Q-Detection also presents Q-WAN optimized using quantum computing devices. Experiments using several quantum simulation libraries show that Q-Detection effectively defends against label manipulation and backdoor attacks. The metrics show that Q-Detection consistently outperforms baseline methods and is comparable to state-of-the-art methods. According to theoretical analysis, Q-Detection is expected to achieve speedups of more than 20% by leveraging quantum computing power.