A team of researchers from the University of Texas at Dallas has developed an approach with an industry collaborator to give quantum computers a layer of protection against adversarial attacks.
The solution, Quantum Noise Injection for Adversarial Defence (QNAD), counteracts the impact of adversarial attacks designed to disrupt the interference of quantum computers. This is AI’s ability to make decisions or solve tasks.
“Adversarial attacks designed to disrupt AI inference have the potential for serious consequences,” said Dr Kanad Basu, assistant professor of electrical and computer engineering at the Erik Jonsson School of Engineering and Computer Science.
The work will be presented at the IEEE International Symposium on Hardware Oriented Security and Trust on 6-9 May in Washington, DC.
Benefits of quantum computers
Quantum computers can solve several complex problems exponentially faster than classical computers. The emerging technology uses quantum mechanics and is expected to improve AI applications and solve complex computational problems.
Qubits represent the fundamental unit of information in quantum computers, like bits in traditional computers.
In classical computers, bits represent 1 or 0. However, qubits take advantage of the principle of superposition and can, therefore, be in a state of 0 and 1. By representing two states, quantum computers have greater speed compared to traditional computers.
For example, quantum computers have the potential to break highly secure encryption systems due to their computer power.
Challenges of quantum computers
Despite their advantages, quantum computers are vulnerable to adversarial attacks.
Due to factors such as temperature fluctuations, magnetic fields, and imperfections in hardware components, quantum computers are susceptible to noise or interference.
Quantum computers are also prone to unintended interactions between qubits.
These challenges can cause computing errors.
Leveraging quantum noise
The researchers leveraged intrinsic quantum noise and crosstalk to counteract adversarial attacks.
The method introduced crosstalk into the quantum neural network. This is a form of Machine Learning where datasets train computers to perform tasks. This includes detecting objects like stop signs or other computer vision responsibilities.
“The noisy behaviour of quantum computers actually reduces the impact of attacks,” said Basu, who is senior author of the study. “We believe this is a first-of-its-kind approach that can supplement other defences against adversarial attacks.”
AI application 268% more accurate with QNAD
The researchers revealed that during an adversarial attack, the AI application was 268% more accurate with QNAD than without it.
The approach is designed to supplement other techniques to protect quantum computer security.
“In case of a crash, if we do not wear the seat belt, the impact of the accident is much greater,” Shamik Kundu, a computer engineering doctoral student and a first co-author, said.
“On the other hand, if we wear the seat belt, even if there is an accident, the impact of the crash is lessened. The QNAD framework operates akin to a seat belt, diminishing the impact of adversarial attacks, which symbolise the accident, for a QNN model.”
The research was funded by the National Science Foundation.
