Quantum generative models offer a novel approach to exploring high-dimensional Hilbert spaces but face significant challenges in scalability and expressibility when applied to multi-modal distributions. In this study, we explore a Hybrid Quantum-Classical U-Net architecture integrated with Adaptive Non-Local Observables (ANO) as a potential solution to these hurdles. By compressing classical data into a dense quantum latent space and utilizing trainable observables, our model aims to extract non-local features that complement classical processing. We also investigate the role of Skip Connections in preserving semantic information during the reverse diffusion process. Experimental results on the full MNIST dataset (digits 0-9) demonstrate that the proposed architecture is capable of generating structurally coherent and recognizable images for all digit classes. While hardware constraints still impose limitations on resolution, our findings suggest that hybrid architectures with adaptive measurements provide a feasible pathway for mitigating mode collapse and enhancing generative capabilities in the NISQ era.
Quantum generative models offer a novel approach to exploring high-dimensional Hilbert spaces but face significant challenges in scalability and expressibility when applied to multi-modal distributions. In this study, we explore a Hybrid Quantum-Classical U-Net architecture integrated with Adaptive Non-Local Observables (ANO) as a potential solution to these hurdles. By compressing classical data into a dense quantum latent space and utilizing trainable observables, our model aims to extract non-local features that complement classical processing. We also investigate the role of Skip Connections in preserving semantic information during the reverse diffusion process. Experimental results on the full MNIST dataset (digits 0-9) demonstrate that the proposed architecture is capable of generating structurally coherent and recognizable images for all digit classes. While hardware constraints still impose limitations on resolution, our findings suggest that hybrid architectures with adaptive measurements provide a feasible pathway for mitigating mode collapse and enhancing generative capabilities in the NISQ era.
Abstract Quantum generative models offer a novel approach to exploring high-dimensional Hilbert spaces but face significant challenges in scalability and expressibility, particularly when applied to multi-modal distributions. In this study, we propose a \textbf{Hybrid Quantum-Classical U-Net} architecture enhanced by \textbf{Adaptive Non-Local Observables (ANO)} and an \textbf{Ancilla-based Global Feature Extractor}. By compressing classical data into a dense quantum latent space and utilizing trainable observables, our model extracts rich non-local features that complement classical processing. Furthermore, we integrate a Hadamard Test module to capture global structural information, fusing it with dense local features. We also investigate the role of Skip Connections in preserving semantic information during the reverse diffusion process. Experimental results on the full MNIST dataset (digits 0-9) demonstrate that the proposed architecture generates structurally coherent and recognizable images for all digit classes, overcoming the mode collapse observed in prior quantum diffusion models. While hardware constraints necessitate resolution downscaling, our findings suggest that hybrid architectures with adaptive measurements provide a feasible pathway for enhancing generative capabilities in the NISQ era.
Quantum generative models offer a promising new direction in machine learning by leveraging quantum circuits to enhance data generation capabilities. In this study, we propose a hybrid quantum-classical image generation framework that integrates variational quantum circuits into a diffusion-based model. To improve training dynamics and generation quality, we introduce two novel noise strategies: intrinsic quantum-generated noise and a tailored noise scheduling mechanism. Our method is built upon a lightweight U-Net architecture, with the quantum layer embedded in the bottleneck module to isolate its effect. We evaluate our model on MNIST and MedMNIST datasets to examine its feasibility and performance. Notably, our results reveal that under limited data conditions (fewer than 100 training images), the quantum-enhanced model generates images with higher perceptual quality and distributional similarity than its classical counterpart using the same architecture. While the quantum model shows advantages on grayscale data such as MNIST, its performance is more nuanced on complex, color-rich datasets like PathMNIST. These findings highlight both the potential and current limitations of quantum generative models and lay the groundwork for future developments in low-resource and biomedical image generation.
The rapid adoption of diffusion models (DMs) in the Earth Observation (EO) domain has unlocked new generative capabilities aimed at producing new samples, whose statistical properties closely match real imagery, for tasks such as synthesizing missing data, augmenting scarce labeled datasets, and improving image reconstruction. This is particularly relevant in EO, where labeled data are often costly to obtain and limited in availability. However, classical DMs still face significant computational limitations, requiring hundreds to thousands of inference steps, as well as difficulties in capturing the intricate spatial and spectral correlations characteristic of EO data. Recent research in Quantum Machine Learning (QML), including initial attempts of Quantum Generative Models, offers a fundamentally different approach to overcome these challenges. Motivated by these considerations, we introduce the Quanvolutional Conditioned U-Net (QCU-Net), a hybrid quantum--classical architecture that applies quantum operations within a conditioned diffusion framework using a novel quanvolutional feature-extraction approach, for generating synthetic labeled EO imagery. Extensive experiments on the EuroSAT RGB dataset demonstrate that our QCU-Net achieves superior results. Notably, it reduces the Fréchet Inception Distance by 64%, lowers the Kernel Inception Distance by 76%, and yields higher semantic accuracy. Ablation studies further reveal that strategically positioning quantum layers and employing entangling variational circuits enhance model performance and convergence. This work represents the first successful adaptation of class-conditioned quantum diffusion modeling in the EO domain, paving the way for quantum-enhanced remote sensing imagery synthesis.
Quantum diffusion models are rapidly advancing generative remote-sensing analytics by harnessing quantum computation to deliver higher-quality, more realistic satellite images, while converging faster and with substantially fewer trainable parameters than comparable classical diffusers. In this context, we first realise a fully-quantum latent diffusion model (QLDM) whose denoiser is a variational-quantum circuit acting on a 10-dimensionaI latent code. On EuroSAT, focusing on three land-cover classes Forest, Herbaceous Vegetation and SeaLake, QLDM lowers the Fréchet Inception Distance (FID) and Kernel Inception Distance (KID) by 21.5 % and 29.9 %, respectively, relative to a parameter-matched classical latent diffuser. Yet the severe spatial compression intrinsic to latent processing limits fine-grained detail. To restore image fidelity, we develop a hybrid quantum—classical architecture—the Quanvolutional Conditioned U-Net (QCU-Net)—which inserts entangling quantum layers both at the U-Net bottleneck and as a quanvolutional filter early in the encoder. Trained on the full ten-class EuroSAT RGB set, QCU-Net reaches an FID of 2.57 and a KID of 0.0008, representing 64 % and 76 % improvements over the best classical diffusion baseline (FID = 7.22; KID = 0.0034), and boosts class-conditioning accuracy to 81.7 % versus 62.2 % for the classical model. Thèse gains confirm that embedding quantum circuits within the feature-extraction pipeline yields richer spatial— spectral representations than latent-space quantum processing alone. Our results outline a clear progression in quantum generative modelling for Earth Observation: by first demonstrating with the QLDM that fully quantum denoising is not only feasible but also highly parameter- efficient, then showing how the hybrid QCU-Net architecture restores perceptual fidelity by embedding quantum feature extractors within a classical U-Net backbone, and finally pointing toward a fully pixel-space quantum diffusion model as future work—one that would leverage deep quantum circuits to perform both the forward and reverse diffusion directly in Hilbert space, remove classical auto-encoding steps, and promise even greater image quality and efficiency on fault-tolerant hardware.
Quantum diffusion models provide a physics-consistent route to generative learning by formulating noising and denoising directly on quantum states. However, applying such models to classical high-dimensional data is constrained by the qubit cost of state encoding and the computational burden of simulating large density operators. We propose a scalable hybrid generative pipeline that combines a classical autoencoder for dimensionality reduction with a mixed-state quantum denoising diffusion probabilistic model (MSQuDDPM) operating in the learned latent space. The autoencoder compresses data into compact latent codes that can be embedded into a small-qubit Hilbert space, after which the quantum diffusion model learns a generative distribution over latent density operators and decodes samples back to the original domain. Algorithmically, we simplify the reverse dynamics by predicting an estimate of the clean state ρ0 at timestep t and computing the one-step reverse update via an analytic backward propagation rule, rather than learning an explicit predictor for ρt−1. We demonstrate the proposed approach on MNIST image generation and discuss how mixed-state quantum diffusion can serve as a practical backbone for hybrid quantum--classical generative modeling under realistic qubit budgets.
The development of compact and controllable generative models under resource constraints remains a critical challenge for the practical deployment of conditional generative adversarial networks (cGANs). In this work, QuCon-GAN, a hybrid quantum-classical conditional GAN, is proposed to enhance generation expressiveness while drastically reducing parameter counts. Within the generator, dynamic data re-upload, parameterized unitary transformations, and adaptive entanglement have been incorporated into a variational quantum circuit, thereby enabling continuous semantic guidance and feature refinement across layers. When compared to classical counterparts, QuCon-GAN achieves significant improvements in parameter efficiency and conditional control, supporting full-category generation on the MNIST dataset at a resolution of 32×32 pixels. Extensive experiments have demonstrated that QuCon-GAN attains a Fréchet Inception Distance (FID) of 52.42, while maintaining stable adversarial training and avoiding mode collapse even under limited quantum resources (8 qubits). Furthermore, systematic comparisons against classical baselines and recent quantum GANs have validated the effectiveness of QuCon-GAN in balancing resource efficiency, stability, and semantic controllability. This work demonstrates the potential of hybrid quantum-classical architectures for advancing lightweight generative modeling in the Noisy Intermediate-Scale Quantum (NISQ) era.