Quantum neural networks (QNN) hold immense potential for the future of quantum machine learning (QML). However, QNN security and robustness remain largely unexplored. In this work, we proposed novel Trojan attacks based on the quantum computing properties in a QNN-based binary classifier. Our proposed Quantum Properties Trojans (QuPTs) are based on the unitary property of quantum gates to insert noise and Hadamard gates to enable superposition to develop Trojans and attack QNNs. We showed that the proposed QuPTs are significantly stealthier and heavily impact the quantum circuits’ performance, specifically QNNs. The most impactful QuPT caused a deterioration of $23 \%$ accuracy of the compromised QNN under the experimental setup. To the best of our knowledge, this is the first work on the Trojan attack on a fully quantum neural network independent of any hybrid classical-quantum architecture.
Quantum neural networks (QNN) hold immense potential for the future of quantum machine learning (QML). However, QNN security and robustness remain largely unexplored. In this work, we proposed novel Trojan attacks based on the quantum computing properties in a QNN-based binary classifier. Our proposed Quantum Properties Trojans (QuPTs) are based on the unitary property of quantum gates to insert noise and Hadamard gates to enable superposition to develop Trojans and attack QNNs. We showed that the proposed QuPTs are significantly stealthier and heavily impact the quantum circuits’ performance, specifically QNNs. The most impactful QuPT caused a deterioration of 23% accuracy of the compromised QNN under the experimental setup. To the best of our knowledge, this is the first work on the Trojan attack on a fully quantum neural network independent of any hybrid classical-quantum architecture.
Quantum computing holds tremendous potential for various applications, but its security remains a crucial concern. Quantum circuits need high-quality compilers to optimize the depth and gate count to boost the success probability on current noisy quantum computers. There is a rise of efficient but unreliable/untrusted compilers; however, they present a risk of tampering such as Trojan insertion. We propose TrojanNet, a novel approach to enhance the security of quantum circuits by detecting and classifying Trojan-inserted circuits. In particular, we focus on the Quantum Approximate Optimization Algorithm (QAOA) circuit that is popular in solving a wide range of optimization problems. We investigate the impact of Trojan insertion on QAOA circuits and develop a Convolutional Neural Network (CNN) model, referred to as TrojanNet, to identify their presence accurately. Using the Qiskit framework, we generate 12 diverse datasets by introducing variations in Trojan gate types, the number of gates, insertion locations, and compiler backends. These datasets consist of both original Trojan-free QAOA circuits and their corresponding Trojan-inserted counterparts. The generated datasets are then utilized for training and evaluating the TrojanNet model. Experimental results showcase an average accuracy of 98.80% and an average F1-score of 98.53% in effectively detecting and classifying Trojan-inserted QAOA circuits. Finally, we conduct a performance comparison between TrojanNet and existing machine learning-based Trojan detection methods specifically designed for conventional netlists.
Quantum computing holds tremendous potential for various applications, but its security remains a crucial concern. Quantum circuits need high-quality compilers to optimize the depth and gate count to boost the success probability on current noisy quantum computers. There is a rise of efficient but unreliable/untrusted compilers; however, they present a risk of tampering, such as malicious insertion of Trojans, which can degrade circuit performance and reliability. This work focuses on analyzing the impact of Trojans in Quantum Approximate Optimization Algorithm (QAOA) circuits, which are widely used for solving combinatorial optimization problems. We propose a methodology to reveal vulnerable locations and adversarial gate types for Trojan insertion that maximizes the negative impact on QAOA’s approximation ratio in solving Max-Cut problem. By disrupting critical paths and altering qubit states, the strategic insertion of additional gates degrade the approximation ratio by up to 50% based on evaluations on benchmark graphs. These insights on plausible attack mechanisms advance the understanding of optimization-oriented Trojan vulnerabilities specific to quantum computing. Additionally, a Convolutional Neural Network (CNN) model, referred to as QTrojanNet, is presented to detect the presence of Trojans in compiled QAOA circuits by learning inherent features that indicate malicious modifications. Experimental results showcase an average accuracy of 98.80% and an average F1-score of 98.53% in effectively detecting and classifying Trojan-inserted QAOA circuits.
The escalating threat and impact of network-based attacks necessitate innovative intrusion detection systems. Machine learning has shown promise, with recent strides in Quantum Machine Learning (QML) offering new avenues. However, the potential of quantum computing is tempered by challenges in current noisy intermediate-scale quantum (NISQ) era machines. In this study, we explore Quantum Neural Networks (QNNs) for intrusion detection, optimizing their performance within current quantum computing limitations. Our approach includes efficient classical feature encoding, QNN classifier selection, and performance tuning leveraging current quantum computational power. This study culminates in an optimized multi-layered QNN architecture for network intrusion detection. A small version of the proposed architecture was implemented on IonQ's Aria-1 quantum computer, achieving a notable 0.86 F1 score using the NF-UNSW-NB15 dataset. Additionally, we introduce a novel metric, certainty factor, laying the foundation for future integration of uncertainty measures in quantum classification outputs. Moreover, this factor is used to predict the noise susceptibility of our quantum binary classification system.
Quantum computing is expected to fundamentally change computer systems in the future. Recently, a new research topic of quantum computing is the hybrid quantum–classical approach for machine learning, in which a parameterized quantum circuit, also called quantum neural network (QNN), is optimized by a classical computer. This hybrid approach can have the benefits of both quantum computing and classical machine learning methods. In this early stage, it is of crucial importance to understand the new characteristics of quantum neural networks for different machine learning tasks. In this paper, we will study quantum neural networks for the task of classifying images, which are high-dimensional spatial data. In contrast to previous evaluations of low-dimensional or scalar data, we will investigate the impacts of practical encoding types, circuit depth, bias term, and readout on classification performance on the popular MNIST image dataset. Various interesting findings on learning behaviors of different QNNs are obtained through experimental results. To the best of our knowledge, this is the first work that considers various QNN aspects for image data.
The reliability of the outcome of a quantum circuit in near-term noisy quantum computers depends on the gate count and depth for a given problem. Circuits with a short depth and lower gate count can yield the correct solution more often than the variant with a higher gate count and depth. To work successfully for Noisy Intermediate Scale Quantum (NISQ) computers, quantum circuits need to be optimized efficiently using a compiler that decomposes high-level gates to native gates of the hardware. Many 3rd party compilers are being developed for lower compilation time, reduced circuit depth, and lower gate count for large quantum circuits. Such compilers, or even a specific release version of a compiler that is otherwise trustworthy, may be unreliable and give rise to security risks such as insertion of a quantum trojan during compilation that evades detection due to the lack of a golden/Oracle model in quantum computing. Trojans may corrupt the functionality to give flipped probabilities of basis states, or result in a lower probability of correct basis states in the output. In this paper, we investigate and discuss the impact of a single qubit Trojan (we have chosen a Hadamard gate and a NOT gate) inserted one at a time at various locations in benchmark quantum circuits without changing the the depth of the circuit. Results indicate an average of 16.18% degradation for the Hadamard Trojan without noise, and 7.78% with noise. For the NOT Trojan (with noise) there is 14.6% degradation over all possible inputs. We then discuss the detection of such Trojans in a quantum circuit using CNN-based classifier achieving an accuracy of 90%.
Quantum machine learning has emerged as a promising application domain for near-term quantum hardware, particularly through hybrid quantum-classical models that leverage both classical and quantum processing. Although numerous hybrid architectures have been proposed and demonstrated successfully on benchmark tasks, a significant open question remains regarding the specific contribution of quantum components to the overall performance of these models. In this work, we aim to shed light on the impact of quantum processing within hybrid quantum-classical neural network architectures through a rigorous statistical study under ideal, noise-free simulation. We systematically assess a selection of commonly used HQNN architectures on three binary medical classification tasks covering medical signal data as well as planar and volumetric images. In doing so, we examine the influence attributable to classical and quantum aspects such as encoding schemes, entanglement, and circuit size. Within this experimental setting, we find that in best-case scenarios, hybrid models show performance comparable to their classical counterparts. However, in most cases, performance metrics deteriorate when the investigated QNN components replace matched classical processing layers. Our multi-modal analysis provides empirical insights into the contributions of the tested quantum components and advocates for cautious claims and design choices for hybrid models in near-term applications.