Quantum Generative Adversarial Networks (qGANs) are leading image-generating quantum machine learning models. With the increasing demand for Noisy Intermediate-Scale Quantum (NISQ) devices, more third-party vendors are expected to offer quantum hardware as a service, raising the risk of proprietary information theft. To mitigate this, we propose a novel watermarking technique that uses the noise signature embedded during qGAN training as a non-invasive watermark. This watermark is detectable in the generated images, tracing the specific quantum hardware used for training and providing strong proof of ownership. To enhance security, we propose training qGANs on multiple quantum hardware, embedding a complex watermark comprising the noise signatures of all training hardware, making it difficult for adversaries to replicate. A machine learning classifier is developed to extract this watermark, identifying the training hardware from the generated images, thereby validating the authenticity of the model. The watermark signature is robust even when inference occurs on different hardware. We achieve watermark extraction accuracy of 100% for single-hardware training and ~ 90% for multi-hardware training setups. We also obtain a validation accuracy of ~ 90% on both single-hardware and multi-hardware training setups in the presence of temporal variation of noise. This watermarking method can be extended to other quantum machine learning models due to the strong modulation of parameter evolution by quantum noise during training.
Quantum Generative Adversarial Networks (qGANs) are leading image-generating quantum machine learning models. With the increasing demand for Noisy Intermediate-Scale Quantum (NISQ) devices, more third-party vendors are expected to offer quantum hardware as a service, raising the risk of proprietary information theft. To mitigate this, we propose a novel watermarking technique that uses the noise signature embedded during qGAN training as a non-invasive watermark. This watermark is detectable in the generated images, tracing the specific quantum hardware used for training and providing strong proof of ownership. To enhance security, we propose training qGANs on multiple quantum hardware, embedding a complex watermark comprising the noise signatures of all training hardware, making it difficult for adversaries to replicate. A machine learning classifier is developed to extract this watermark, identifying the training hardware from the generated images, thereby validating the authenticity of the model. The watermark signature is robust even when inference occurs on different hardware. We achieve watermark extraction accuracy of 100% for single-hardware training and ~ 90% for multi-hardware training setups. We also obtain a validation accuracy of ~ 90% on both single-hardware and multi-hardware training setups in the presence of temporal variation of noise. This watermarking method can be extended to other quantum machine learning models due to the strong modulation of parameter evolution by quantum noise during training.
Machine learning can be used as a systematic method to non-algorithmically "program" quantum computers. Quantum machine learning enables us to perform computations without breaking down an algorithm into its gate "building blocks", eliminating that difficult step and potentially reducing unnecessary complexity. In addition, the machine learning approach is robust to both noise and to decoherence, which is ideal for running on inherently noisy NISQ devices which are limited in the number of qubits available for error correction. Here we apply our prior work in quantum machine learning technique, to create a QGAN, a quantum analog to the classical Stylenet GANs developed by Kerras for image generation and classification. A quantum system is used as a generator and a separate quantum system is used as a discriminator. The generator Hamiltonian quantum parameters are augmented by quantum "style" parameters which play the role of the style parameters used by Kerras. Both generator parameters are trained in a GAN MinMax problem along with quantum parameters of the discriminator Hamiltonian. We choose a problem to demonstrate the QGAN that has purely quantum information. The task is to generate and discriminate quantum product states. "Real" product states are generated to be detected and separated from "fake" quantum product states generated by the quantum generator. The problem of detecting quantum product states is chosen to demonstrate the QGAN because is well known as purely quantum mechanical, has no classical analog and is an open problem for quantum product states of more than 2 qubits. With proper encoding of image pixels into quantum states as density matrices, the method demonstrated here for this quantum problem is applicable to GAN image generation and detection, the problem so successfully demonstrated by Kerras's classical GAN, but can be hosted on and taking advantage of the nature of quantum computers.
Machine learning can be used as a systematic method to non-algorithmically program quantum computers. Quantum machine learning enables us to perform computations without breaking down an algorithm into its gate building blocks, eliminating that difficult step and potentially reducing unnecessary complexity. In addition, the machine learning approach is robust to both noise and to decoherence, which is ideal for running on inherently noisy NISQ devices which are limited in the number of qubits available for error correction. Here we apply our prior work in quantum machine learning technique, to create a QGAN, a quantum analog to the classical Stylenet GANs developed by Kerras for image generation and classification. A quantum system is used as a generator and a separate quantum system is used as a discriminator. The generator Hamiltonian quantum parameters are augmented by quantum style parameters which play the role of the style parameters used by Kerras. Both generator parameters are trained in a GAN MinMax problem along with quantum parameters of the discriminator Hamiltonian. We choose a problem to demonstrate the QGAN that has purely quantum information. The task is to generate and discriminate quantum product states. Real product states are generated to be detected and separated from fake quantum product states generated by the quantum generator. The problem of detecting quantum product states is chosen to demonstrate the QGAN because is well known as purely quantum mechanical, has no classical analog and is an open problem for quantum product states of more than 2 qubits. With proper encoding of image pixels into quantum states as density matrices, the method demonstrated here is applicable to GAN image generation and detection that can be hosted on and take advantage of the nature of quantum computers.
Quantum Machine Learning (QML) has attracted significant attention for its potential to deliver exponential advantages over classical machine learning approaches, particularly in classification and recognition tasks. Quantum Generative Adversarial Networks (QGANs), a form of quantum machine learning, provide promising advantages in image processing and generation tasks when compared to classical technologies. However, the limitations of current quantum devices have led to suboptimal image quality and limited robustness in earlier methods. To overcome these challenges, we developed a hybrid quantum-classical approach, introducing CAQ, a quantum-classical Generative Adversarial Network (GAN) framework. Leveraging the latest WGAN-gradient penalty (GP) strategy, we trained and optimized the quantum generator, reduced the complexity of parameters, and implemented an adaptive noise input system that dynamically adjusts noise levels, thereby improving the model’s robustness. Additionally, we employed a remapping technique to transform the original image’s multimodal distribution into a unimodal one, thereby reducing the complexity of the learned distribution. Experiments on MNIST and Fashion-MNIST datasets show that CAQ generates grayscale images effectively, demonstrating its feasibility on near-term intermediate-scale quantum (NISQ) computers.
Quantum Generative Adversarial Networks (QGANs) represent a useful development in quantum machine learning, using the particular properties of quantum mechanics to solve the challenges of data analysis and modeling. This paper brings up a general analysis of five QGAN architectures, focusing on their evolution, strengths, weaknesses, and limitations in noisy intermediate-scale quantum (NISQ) devices. Primary methods like Entangling Quantum GAN (EQ-GAN) and Quantum state fidelity (QuGAN) concentrate on stability, convergence, and robust performance on small-scale datasets such as 2 × 2 grayscale images. Intermediate models such as Image Quantum GAN (IQGAN) and Experimental Quantum GAN (EXQGAN) provide new ideas like trainable encoders and patch-based sub-generators that are scalable to 8 × 8 datasets with increasing noise resilience. The most advanced method is Parameterized Quantum Wasserstein GAN (PQWGAN), which uses a hybrid quantum-classical structure to obtain high-resolution image processing for 28 × 28 grayscale datasets while trying to maintain parameter efficiency. This study explores, analyzes, and summarizes critical problems of QGANs, including accuracy, convergence, parameter efficiency, image quality, performance metrics, and training stability under noisy conditions. In addition, developing QGANs can generate and train parameters in quantum approximation optimization algorithms. One of the useful applications of QGAN is generating medical datasets that can generate medical images from limited datasets to train specific medical models for the recognition of diseases.
This paper introduces a novel hybrid quantumclassical model using QGAN (quantum generative adversarial network) to achieve more efficient, convergent and noiserobust machine learning for nearterm quantum (NISQ) applications. Unlike the existing GAN type models that were purely classical or purely quantum, to propose a quantum generator consisting of shallow variation circuits (48 qubits) together with a classical discriminator that is crucial for the stability of adversarial training. The empirical results on the MNIST dataset showed that were able to perform classification with an accuracy of 94.2%, an improvement of over 3.2% from a straightforward classical only approach, while NVGAN also improved its generative performance to a Fréchet Inception Distance (FID) score of 38.7 or 26% in quality enhancement from synthesis perspective. Both trainings converged 50 epochs earlier than each training when used the hybrid model (30% improvement in convergence). The key innovation of this work is bringing (1) to bear both the utility of the quantumenhanced feature representation to have at disposal, and (2) classical stability measures and erroraware optimizations (such as parameter clipping, early stopping), such that its hybridization enabled it to continue operating with>90%accuracy under cases of reasonable depolarizing/amplitude damping noise on an MLP where purely quantum formulations decayed to (approx 70%). Altogether, the paper validates the presented hybrid QGAN approach as a statistically substantiated noise robust and scalable representational tool for real world applications of quantum enhanced machine learning tasks on NISQ platforms.
The potential advantage of machine learning in quantum computers is a topic of intense discussion in the literature. Theoretical, numerical and experimental explorations will most likely be required to understand its power. There has been different algorithms proposed to exploit the probabilistic nature of variational quantum circuits for generative modelling. In this paper, we employ a hybrid architecture for quantum generative adversarial networks (QGANs) and study their robustness in the presence of noise. We devise a simple way of adding different types of noise to the quantum generator circuit, and numerically simulate the noisy hybrid QGANs to learn continuous probability distributions, and show that the performance of HQGANs remain unaffected. We also investigate the effect of different parameters on the training time to reduce the computational scaling of the algorithm and simplify its deployment on a quantum computer. We then perform the training on Rigetti's Aspen-4-2Q-A quantum processing unit, and present the results from the training. Our results pave the way for experimental exploration of different quantum machine learning algorithms on noisy intermediate scale quantum devices.