While Quantum Convolutional Neural Networks (QCNNs) offer a theoretical paradigm for quantum machine learning, their practical implementation is severely bottlenecked by barren plateaus -- the exponential vanishing of gradients -- and poor empirical accuracy compared to classical counterparts. In this work, we propose a novel QCNN architecture utilizing localized cost functions and a hardware-efficient tensor-network initialization strategy to provably mitigate barren plateaus. We evaluate our scalable QCNN on the MNIST dataset, demonstrating a significant performance leap. By resolving the gradient vanishing issue, our optimized QCNN achieves a classification accuracy of 98.7\%, a substantial improvement over the baseline QCNN accuracy of 52.32\% found in unmitigated models. Furthermore, we provide empirical evidence of a parameter-efficiency advantage, requiring $\mathcal{O}(\log N)$ fewer trainable parameters than equivalent classical CNNs to achieve $>95\%$ convergence. This work bridges the gap between theoretical quantum utility and practical application, providing a scalable framework for quantum computer vision tasks without succumbing to loss landscape concentration.
While Quantum Convolutional Neural Networks (QCNNs) offer a theoretical paradigm for quantum machine learning, their practical implementation is severely bottlenecked by barren plateaus -- the exponential vanishing of gradients -- and poor empirical accuracy compared to classical counterparts. In this work, we propose a novel QCNN architecture utilizing localized cost functions and a hardware-efficient tensor-network initialization strategy to provably mitigate barren plateaus. We evaluate our scalable QCNN on the MNIST dataset, demonstrating a significant performance leap. By resolving the gradient vanishing issue, our optimized QCNN achieves a classification accuracy of 98.7%, a substantial improvement over the baseline QCNN accuracy of 52.32% found in unmitigated models. Furthermore, we provide empirical evidence of a parameter-efficiency advantage, requiring O(logN) fewer trainable parameters than equivalent classical CNNs to achieve >95% convergence. This work bridges the gap between theoretical quantum utility and practical application, providing a scalable framework for quantum computer vision tasks without succumbing to loss landscape concentration.
Abstract Quantum Neural Networks (QNNs) present a promising research direction in image recognition, demonstrating significant potential for various applications. However, QNNs with a fixed circuit topology structure are prone to encountering the barren plateaus during training when processing exponential number of qubits, which limit their performance and scalability. In this paper, a novel Hybrid Quantum-Classical Inception-inspired Convolutional Neural Networks (HQCNN) is proposed. The HQCNN enhances feature learning capability and mitigates the barren plateau issue by integrating quantum convolutional kernels with diverse circuit topologies across multiple channels. The architecture employs an amplitude encoding layer to map multi-channel classical data into quantum states, followed by diverse quantum convolutional kernels that extract feature representations. Furthermore, Hadamard gates are used to intergrate multi-channel features and incorporate quantum pooling to reduce feature dimensionality, with a fully connected layer mapping the quantum-enhancing features into class labels for classification tasks. Finally, by analyzing the influence of the circuit structures and quantum gate combinations on the models' performance, the optimal HQCNN architecture is constructed. Experimental results demonstrate that HQCNN achieves 98.5% accuracy on the MNIST-4 classification task and 92.5% accuracy on the MNIST-10 classification task, which indicates that the proposed model outperforms state-of-the-art HQCNN models and classical CNN in image classification with about 10% parameters. This study provides an effective approach for alleviating the barren plateau issue and designing multi-channel quantum convolutional neural networks for processing big classic data with fewer quantum resources.
Quantum neural networks (QNNs) have generated excitement around the possibility of efficiently analyzing quantum data.But this excitement has been tempered by the existence of exponentially vanishing gradients, known as barren plateau landscapes, for many QNN architectures.Recently, quantum convolutional neural networks (QCNNs) have been proposed, involving a sequence of convolutional and pooling layers that reduce the number of qubits while preserving information about relevant data features.In this work, we rigorously analyze the gradient scaling for the parameters in the QCNN architecture.We find that the variance of the gradient vanishes no faster than polynomially, implying that QCNNs do not exhibit barren plateaus.This result provides an analytical guarantee for the trainability of randomly initialized QCNNs, which highlights QCNNs as being trainable under random initialization unlike many other QNN architectures.To derive our results, we introduce a novel graph-based method to analyze expectation values over Haar-distributed unitaries, which will likely be useful in other contexts.Finally, we perform numerical simulations to verify our analytical results.
Quantum neural networks (QNNs) leverage quantum entanglement and superposition to enable large-scale parallel linear computation, offering a potential solution to the scalability limits of classical deep learning. However, their practical deployment is hampered by two key challenges: the lack of intrinsic nonlinear operations and the barren plateau phenomenon. We propose a quantum convolutional neural network (QCNN) architecture that simultaneously addresses both issues. Nonlinear effects are introduced via orthonormal basis expansions of power series, while barren plateaus are mitigated by directly parameterizing unitary matrices rather than stacking multiple parameterized gates. Our design further incorporates quantum analogs of convolutional kernels and strides for scalable circuit construction. Experiments on MNIST and Fashion-MNIST datasets achieve 99.0% and 88.0% test accuracy, respectively. Consistency between PyTorch-based matrix simulation and Qiskit-based quantum circuit simulation validates the physical fidelity of the model. These results demonstrate a flexible and effective quantum architecture that faithfully integrates classical convolutional mechanisms into a quantum framework, paving the way for practical and expressive QNNs.
Realizing successful quantum algorithms on current generation of quantum hardware depends on developing appropriate parametrized quantum circuits, which are the quantum computing counterpart to classical neural networks. A key consideration in this context is the algorithm’s trainability that can be hindered by the existence of so-called barren plateaus. In this case, the algorithm suffers from exponentially vanishing gradients as the number of qubits increases, resulting in a flat optimization landscape and making an efficient training intractable. Due to its serious impact on the success of such algorithms, this problem has gained substantial attention in recent years of quantum computing research. Theoretical and heuristic approaches have been proposed to investigate, avoid and mitigate this issue. In this work, we explore this phenomenon in detail and study the underlying mathematical concepts in depth, investigating the factors that influence the structure of the optimization landscape. Using corresponding mathematical tools, we diagnose whether a given variational circuit will exhibit barren plateaus when going beyond tens of qubits. Additionally, a review of mitigation strategies will be provided. The theory will then be applied to a use case that involves image classification to evaluate whether the model is affected by this phenomenon and which hyperparameters improve or degrade the overall performance.
Convolutional Neural Networks (CNNs) are extensively utilized for image classification due to their ability to exploit data correlations effectively. However, traditional CNNs encounter several challenges, including limitations in accuracy improvement, computational efficiency, explain ability, and generalization. As data dimensionality or model complexity increases, CNNs often become more difficult to train efficiently, leading to slower processing times. To overcome these challenges, Quantum Convolutional Neural Networks (QCNNs) offer a promising quantum-based approach to enhance existing models, particularly by improving training efficiency and reducing processing times. This paper provides a comparative analysis of classical CNNs and a novel quantum circuit architecture tailored for image-based tasks. The analysis emphasizes the adaptability and potential of quantum circuits to enhance feature extraction, recognition accuracy, and processing efficiency.We applied various QCNN architectures to the Chinese-MNIST dataset and compared their performance with traditional CNN models. Although QCNNs achieved slightly lower accuracy, they demonstrated significantly greater stability. These findings high- light QCNNs’ potential to overcome the computational limitations of conventional deep learning, offering a promising pathway for quantum-enhanced solutions in computer vision. Future work will focus on refining quantum circuit designs, developing hybrid quantum-classical models, and extending QCNN applications to larger and more diverse datasets.
Abstract Quantum Convolutional Neural Networks (QCNNs) are a significant quantum machine learning architecture. However, most current designs still mimic the ‘convolution-pooling’ structure of classical CNNs, facing dual challenges of limited feature extraction and inefficient quantum resource usage. To address this, we propose a QCNN based on a hierarchical Ansatz design. Its core is an alternating ‘convolution-entanglement’ structure, which replaces abrupt qubit discarding in traditional pooling with sequential global entanglement layers. This enables non-destructive directional information aggregation, aiming to preserve quantum correlations in data more comprehensively. In the encoding layer, we integrate Principal Component Analysis (PCA) with Instantaneous Quantum Polynomial (IQP) encoding to mitigate the dimensionality mismatch between high-dimensional classical data and limited quantum resources. To systematically evaluate the performance of the proposed architecture, we conducted extensive experiments on the MNIST and Fashion-MNIST datasets. The results demonstrate that: 1) Under various convolutional kernel configurations, our model achieves excellent classification performance, peaking at 99.39% accuracy on MNIST; 2) Even with minimalistic single-qubit rotation gates as the convolutional layer, the model attains 99.14% accuracy using only 16 parameters, demonstrating high parameter efficiency; 3) Crucially, experiments reveal a nonlinear relationship between quantum resources and performance. When qubits exceed a task-specific optimal number, performance declines significantly due to feature space dilution and exacerbated Barren Plateaus. This suggests that, under current quantum hardware conditions, matching resource allocation to task complexity may be more critical than simply scaling up the system. This study offers one potential direction for exploring QCNN architectures. The findings still require further validation across more diverse tasks and in quantum hardware environments closer to practical conditions. We hope this work can provide some useful insights and inform subsequent research.