The field of computer vision has long grappled with the challenging task of image synthesis, which entails the creation of novel high-fidelity images. This task is underscored by the Generative Learning Trilemma, which posits that it is not possible for any image synthesis model to simultaneously excel at high-quality sampling, achieve mode convergence with diverse sample representation, and perform rapid sampling. In this paper, we explore the potential of Quantum Boltzmann Machines (QBMs) for image synthesis, leveraging the D-Wave 2000Q quantum annealer. We undertake a comprehensive performance assessment of QBMs in comparison to established generative models in the field: Restricted Boltzmann Machines (RBMs), Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), and Denoising Diffusion Probabilistic Models (DDPMs). Our evaluation is grounded in widely recognized scoring metrics, including the Fréchet Inception Distance (FID), Kernel Inception Distance (KID), and Inception Scores. The results of our study indicate that QBMs do not significantly outperform the conventional models in terms of the three evaluative criteria. Moreover, QBMs have not demonstrated the capability to overcome the challenges outlined in the Trilemma of Generative Learning. Through our investigation, we contribute to the understanding of quantum computing’s role in generative learning and identify critical areas for future research to enhance the capabilities of image synthesis models.
The field of computer vision has long grappled with the challenging task of image synthesis, which entails the creation of novel high-fidelity images. This task is underscored by the Generative Learning Trilemma, which posits that it is not possible for any image synthesis model to simultaneously excel at high-quality sampling, achieve mode convergence with diverse sample representation, and perform rapid sampling. In this paper, we explore the potential of Quantum Boltzmann Machines (QBMs) for image synthesis, leveraging the D-Wave 2000Q quantum annealer. We undertake a comprehensive performance assessment of QBMs in comparison to established generative models in the field: Restricted Boltzmann Machines (RBMs), Variational Autoencoders (VAEs), Generative Adversarial Networks (GANs), and Denoising Diffusion Probabilistic Models (DDPMs). Our evaluation is grounded in widely recognized scoring metrics, including the Fréchet Inception Distance (FID), Kernel Inception Distance (KID), and Inception Scores. The results of our study indicate that QBMs do not significantly outperform the conventional models in terms of the three evaluative criteria. Moreover, QBMs have not demonstrated the capability to overcome the challenges outlined in the Trilemma of Generative Learning. Through our investigation, we contribute to the understanding of quantum computing’s role in generative learning and identify critical areas for future research to enhance the capabilities of image synthesis models.
Quantum Generative Adversarial Networks (QGANs), as a rising paradigm in Quantum Machine Learning, have shown promising potential in image generation and processing. However, their output quality remains suboptimal, and existing research is largely limited to small-scale, proof-of-concept studies. In this work, we propose a hybrid quantum-classical GAN architecture, where the generator integrates parameterized quantum circuits (PQCs) and classical neural networks. This integration significantly enhances the visual quality of generated images. Our model leverages the complementary strengths of quantum and classical components and outperforms existing methods (Tsang et al. in IEEE Trans. Quantum Eng. 4:1–19, 2023 ; Gulrajani et al. in Proceedings of the 31st international conference on neural information processing systems. NIPS’17, Red Hook, pp. 5769–5779, 2017 ), particularly in terms of image fidelity. Experiments conducted on the MNIST family of datasets show that our hybrid approach achieves a 20.26% average reduction in Fréchet Inception Distance. Furthermore, it improves the Structural Similarity Index Measure, Cosine Similarity, and Peak Signal-to-Noise Ratio by 26.04%, 2.22%, and 7.62%, respectively. These results highlight the effectiveness of combining quantum computing with machine learning, and underscore the potential of hybrid quantum-classical models in advancing generative tasks.
Generative Adversarial Networks (GANs) have demonstrated immense potential in synthesizing diverse and high-fidelity images; however, critical questions remain regarding how quantum principles might best enhance their representational capacity. This thesis investigates hybrid quantum-classical GAN architectures supplemented by transfer learning in the discriminator, systematically examining whether incorporating Variational Quantum Circuits (VQCs) into the generator, the discriminator, or both improves performance over fully classical baselines. We evaluate each configuration using standard GAN benchmarks (CIFAR-10 classes: Birds, Cars, and Dogs) under varying sample sizes (5,000 vs. 2,500 images per class) to gauge both robustness and data efficiency. Our findings indicate that fully hybrid models, which incorporate VQCs in both the generator and the discriminator, consistently produce images of higher visual quality and achieve more favorable quantitative metrics (such as Fréchet Inception Distance (FID), Kernel Inception Distance (KID), and Inception Score (IS)) compared to their fully classical counterpart. In particular, VQCs in the generator accelerate early feature learning, whereas those in the discriminator, despite exhibiting slower initial convergence, ultimately facilitate more refined synthetic outputs. Moreover, the model sustains near-comparable performance even when the dataset size is halved, suggesting that transfer learning and quantum enhancements mitigate data limitations. By comparing classical and hybrid configurations in detail, this thesis illustrates the trade-offs arising from VQCs in GANs, such as variations in early-epoch convergence and final generative quality. Overall, the results underscore that carefully integrating quantum computing principles with classical adversarial training and pretrained feature extraction can significantly enrich GAN-based image synthesis. These insights open avenues for future work on higher-resolution tasks, alternative quantum circuit designs, and experimentation with emerging quantum hardware.
Understanding how the D-Wave quantum computer could be used for machine learning problems is of growing interest. Our work explores the feasibility of using the D-Wave as a sampler for a machine learning task. We describe a hybrid method that combines a classical deep neural network autoencoder with a quantum annealing Restricted Boltzmann Machine (RBM) using the D-Wave for image generation. Our method overcomes two key limitations in the 2000-qubit D-Wave processor, namely the limited number of qubits available to accommodate typical problem sizes for fully connected quantum objective functions, and samples that are binary pixel representations. As a consequence of these limitations we are able to show how we achieved nearly a 22-fold compression factor of grayscale 28 x 28 sized images to binary 6 x 6 sized images with a lossy recovery of the original 28 x 28 grayscale images. We further show how generating samples from the D-Wave after training the RBM, resulted in 28 x 28 images that were variations of the original input data distribution, as opposed to recreating the training samples. We evaluated the quality of this method by using a downstream classification method. We formulated a MNIST classification problem using a deep convolutional neural network that used samples from the quantum RBM to train the MNIST classifier and compared the results with a MNIST classifier trained with the original MNIST training data set, as well as a MNIST classifier trained using classical RBM samples. We also explored using a secondary dataset, the MNIST Fashion dataset and demonstrate the first quantum-generated fashion. Our hybrid autoencoder approach indicates advantage for RBM results relative to the use of a current RBM classical computer implementation for image-based machine learning and even more promising results for the next generation D-Wave quantum system. Our method for compression and image mappings is not constrained to RBMs, the autoencoder part of this method could be coupled with other quantum-based algorithms.
Generative Adversarial Networks (GANs) have demonstrated immense potential in synthesizing diverse and high-fidelity images; however, critical questions remain regarding how quantum principles might best enhance their representational capacity. This thesis investigates hybrid quantum-classical GAN architectures supplemented by transfer learning in the discriminator, systematically examining whether incorporating Variational Quantum Circuits (VQCs) into the generator, the discriminator, or both improves performance over fully classical baselines. We evaluate each configuration using standard GAN benchmarks (CIFAR-10 classes: Birds, Cars, and Dogs) under varying sample sizes (5,000 vs. 2,500 images per class) to gauge both robustness and data efficiency. Our findings indicate that fully hybrid models, which incorporate VQCs in both the generator and the discriminator, consistently produce images of higher visual quality and achieve more favorable quantitative metrics (such as Fréchet Inception Distance (FID), Kernel Inception Distance (KID), and Inception Score (IS)) compared to their fully classical counterpart. In particular, VQCs in the generator accelerate early feature learning, whereas those in the discriminator, despite exhibiting slower initial convergence, ultimately facilitate more refined synthetic outputs. Moreover, the model sustains near-comparable performance even when the dataset size is halved, suggesting that transfer learning and quantum enhancements mitigate data limitations.By comparing classical and hybrid configurations in detail, this thesis illustrates the trade-offs arising from VQCs in GANs, such as variations in early-epoch convergence and final generative quality. Overall, the results underscore that carefully integrating quantum computing principles with classical adversarial training and pretrained feature extraction can significantly enrich GAN-based image synthesis. These insights open avenues for future work on higher-resolution tasks, alternative quantum circuit designs, and experimentation with emerging quantum hardware.
We investigate the quantum effect on machine learning (ML) models exemplified by the Generative Adversarial Network (GAN), which is a promising deep learning framework. In the general GAN framework the generator maps uniform noise to a fake image. In this study, we utilize the Associative Adversarial Network (AAN), which consists of a standard GAN and an associative memory. Further, we set a Boltzmann Machine (BM), which is an undirected graphical model that learns low-dimensional features extracted from a discriminator, as the memory. Owing to difficulty calculating the BM's log-likelihood gradient, it is necessary to approximate it by using the sample mean obtained from the BM, which has tentative parameters. To calculate the sample mean, a Markov Chain Monte Carlo (MCMC) is often used. In a previous study, this was performed using a quantum annealer device, and the performance of the "Quantum" AAN was compared to that of the standard GAN. However, its better performance than the standard GAN is not well understood. In this study, we introduce two methods to draw samples: classical sampling via MCMC and quantum sampling via quantum Monte Carlo (QMC) simulation, which is quantum simulation on the classical computer. Then, we compare these methods to investigate whether quantum sampling is advantageous. Specifically, calculating the discriminator loss, the generator loss, inception score and Fréchet inception distance, we discuss the possibility of AAN. We show that the AANs trained by both MCMC and QMC are more stable during training and produce more varied images than the standard GANs. However, the results indicate no difference in sampling by QMC simulation compared to that by MCMC.
Generative Adversarial Networks (GANs) have demonstrated immense potential in synthesizing diverse and high-fidelity images. However, critical questions remain unanswered regarding how quantum principles might best enhance their representational and computational capacity. In this paper, we investigate hybrid quantum-classical GAN architectures supplemented by transfer learning to systematically examine whether incorporating Variational Quantum Circuits (VQCs) into the generator, the discriminator, or both improves performance over a fully classical baseline. Our findings indicate that fully hybrid models, which incorporate VQCs in both the generator and the discriminator, produce images with higher quality and achieve more favorable quantitative metrics compared to their fully classical counterparts. In particular, placing the quantum block in the generator appears to accelerate the early emergence of visual structure, whereas placing it in the discriminator slows early visual convergence but improves the final quantitative quality metric. Incorporating quantum blocks into both networks yields the strongest overall performance. Moreover, the model sustains comparable performance even when the dataset size is reduced. Overall, the results underscore that carefully integrating quantum computing with classical adversarial training and pretrained feature extraction can enrich GAN-based image synthesis. These insights open avenues for future work on higher-resolution tasks, alternative quantum circuit designs, and experimentation with emerging quantum hardware.