Quantum machine learning with differential privacy.

Area of Science:

• Quantum Computing
• Machine Learning
• Data Privacy

Background:

• Quantum machine learning (QML) offers potential advantages for complex classification tasks.
• Current QML models lack privacy-preserving features, posing risks to sensitive data.
• Differential privacy is a key technique for protecting data in machine learning.

Purpose of the Study:

• To develop and demonstrate a privacy-preserving QML model.
• To investigate the effectiveness of differential privacy in QML.
• To ensure data confidentiality in quantum machine learning applications.

Main Methods:

• Developed a hybrid quantum-classical model.
• Employed a differentially private optimization algorithm for training.
• Tested the model on 2D datasets and MNIST classification.

Main Results:

• The differentially private QML model successfully protected user-sensitive information.
• Model accuracy was maintained without significant degradation.
• The approach is suitable for implementation on noisy intermediate-scale quantum (NISQ) devices.

Conclusions:

• This work presents the first proof-of-principle for privacy-preserving QML.
• Differential privacy can be effectively integrated into QML without compromising performance.
• The developed method ensures confidentiality and accurate learning for NISQ technology.