Quantum-aided secure deep neural network inference on real quantum computers.

Area of Science:

• Quantum Computing
• Machine Learning Security
• Cryptography

Background:

• Deep neural networks (DNNs) are widely used but face security risks when handling sensitive data due to limitations in classical cryptographic methods.
• Achieving unconditional security for DNN inference is crucial for applications involving private or confidential information.

Purpose of the Study:

• To develop a quantum-enhanced security scheme for deep neural network inference.
• To enable secure DNN inference on low-fidelity quantum systems with limited quantum capacity.

Main Methods:

• A novel quantum scheme for unconditionally secure DNN inference was designed, utilizing quantum oblivious transfer with an untrusted third party.
• The approach leverages the inherent noise tolerance of DNNs to operate on existing, imperfect quantum computers.

Main Results:

• The quantum-aided security approach was validated on a five-bit quantum computer and a quantum simulator.
• Experimental results and theoretical analyses confirmed unconditional security with negligible accuracy loss during DNN inference.

Conclusions:

• The developed method successfully integrates quantum security principles with deep learning inference.
• This research opens new avenues for quantum security applications in machine learning, particularly for sensitive data protection.