Training deep quantum neural networks (QNNs) for image classification is notoriously difficult due to vanishing gradients (barren plateaus) and limited nonlinearity in purely unitary circuits. We propose a novel gradient-free surrogate-driven framework combined with mid-circuit measurement and reset of ancillary qubits to induce effective nonunitarity. Our approach uses a classical neural surrogate to predict measurement outcomes from circuit parameters to avoid direct gradients. Theoretical results prove that bypassing quantum gradients mitigates plateau issues. Experiments on MNIST, CIFAR-10, and CIFAR-100 with 15-qubit, 6-layer circuits using four resettable ancillas demonstrate superior accuracy compared to direct-gradient QNNs and classical baselines. Our method also serves as a potential for a generalized training framework applicable to various QNN architectures beyond image classification.
Training deep quantum neural networks (QNNs) for image classification is notoriously difficult due to vanishing gradients (barren plateaus) and limited nonlinearity in purely unitary circuits. We propose a novel gradient-free surrogate-driven framework combined with mid-circuit measurement and reset of ancillary qubits to induce effective nonunitarity. Our approach uses a classical neural surrogate to predict measurement outcomes from circuit parameters to avoid direct gradients. Theoretical results prove that bypassing quantum gradients mitigates plateau issues. Experiments on MNIST, CIFAR-10, and CIFAR-100 with 15-qubit, 6-layer circuits using four resettable ancillas demonstrate superior accuracy compared to direct-gradient QNNs and classical baselines. Our method also serves as a potential for a generalized training framework applicable to various QNN architectures beyond image classification.
Training quantum neural networks (QNNs) on near-term hardware remains hampered by two compounding difficulties: the exponential vanishing of gradient variance known as the barren plateau, and the O(L⋅2n) time and memory cost of differentiating through an n-qubit, L-layer circuit. We propose RLQ-Grad, a reinforcement-learning-based optimizer in which a classical policy πφ (a spectrally-normalized PPO agent) learns to propose parameter updates directly, conditioned on the QNN's current parameters, loss, accuracy, and previous update. Because the surrogate gradient is emitted by a classical network rather than obtained by differentiating through the unitary U(θ), its variance is not constrained by the barren plateau concentration bound, and its cost scales with the number of trainable parameters rather than the Hilbert-space dimension. We prove these properties formally and verify them on a hardware-efficient ansatz across four supervised benchmarks with up to n=20 qubits. RLQ-Grad preserves a near-flat gradient-variance curve where backpropagation, parameter-shift, and adjoint differentiation decay by 1 to 2 orders of magnitude. Accounting for the full training pipeline (PPO rollouts, actor-critic updates, and optimizer states), RLQ-Grad needs under 2 MB of memory and runs 2490×, 7876×, and 673× faster per iteration than these three methods at n=20. It improves top-1 accuracy by up to +10% over gradient-based baselines on circuits of up to 12 qubits, and matches dedicated barren plateau mitigation methods on CIFAR-10 at 14 to 20 qubits, where evolutionary and gradient-free optimizers collapse to chance.
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.
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.
Variational quantum circuits for image classification suffer from barren plateaus, while quantum kernel methods scale quadratically with dataset size. We propose an iterative framework based on Quadratic Unconstrained Binary Optimization (QUBO) for training the classifier head of convolutional neural networks (CNNs) via quantum annealing, entirely avoiding gradient-based circuit optimization. Following the Extreme Learning Machine paradigm, convolutional filters are randomly initialized and frozen, and only the fully connected layer is optimized. At each iteration, a convex quadratic surrogate derived from the feature Gram matrix replaces the non-quadratic cross-entropy loss, yielding an iteration-stable curvature proxy. A per-output decomposition splits the C-class problem into C independent QUBOs, each with (d+1)K binary variables, where d is the feature dimension and K is the bit precision, so that problem size depends on the image resolution and bit precision, not on the number of training samples. We evaluate the method on six image-classification benchmarks (sklearn digits, MNIST, Fashion-MNIST, CIFAR-10, EMNIST, KMNIST). A precision study shows that accuracy improves monotonically with bit resolution, with 10 bits representing a practical minimum for effective optimization; the 15-bit formulation remains within the qubit and coupler limits of current D-Wave Advantage hardware. The 20-bit formulation matches or exceeds classical stochastic gradient descent on MNIST, Fashion-MNIST, and EMNIST, while remaining competitive on CIFAR-10 and KMNIST. All experiments use simulated annealing, establishing a baseline for direct deployment on quantum annealing hardware.
Barren plateaus (BPs) pose a major trainability challenge for quantum neural networks (QNNs) by causing gradients to concentrate near zero as circuit size, depth, or expressibility increases. This study presents a comparative benchmark of 10 BP mitigation strategies across six qubit settings (2, 4, 8, 12, 16, and 20) and three datasets of increasing complexity: Iris, MNIST, and MedMNIST. The evaluated methods include eight initialization-based strategies (Beta, Gaussian, Uniform Norm, CNN-based initialization, He-normal, He-uniform, Xavier-normal, and Xavier-uniform), one model-based variational encoder, and one optimization-based time-nonlocal Fourier parameterization. Experiments were implemented using PennyLane 3.10 and PyTorch 2.5 with simulator backends. We evaluate trainability using gradient variance and training loss, and we clarify that the benchmark analyzes simulated QNN optimization behavior rather than hardware-noise-resilient or noisy-label learning. Across the tested two-layer circuit configurations, the mitigation strategies maintained measurable gradient variance and stable loss reduction, suggesting that severe barren plateau behavior was not observed under the benchmark conditions. CNN-based and Beta initialization showed strong empirical behavior in variance retention and convergence speed, while Gaussian initialization was comparatively weaker in higher-dimensional settings. The study provides a reproducible benchmark structure for comparing BP mitigation behavior and identifies important limitations related to circuit depth, hardware noise, feature encoding, and classification performance that should be addressed in future QNN benchmarking.
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.