Quantum Convolutional Neural Networks Achieve Performance Gains on Downsampled MNIST and Fashion-MNIST

Quantum convolutional neural networks represent a powerful approach to machine learning on near-term quantum computers, but effectively translating classical data into quantum information remains a significant challenge. Xingyun Feng from Northwest University, and colleagues, investigate this crucial step by comparing three leading encoding strategies, Angle, Amplitude, and a Hybrid phase/angle scheme, for quantum convolutional neural networks. Their work demonstrates that the optimal encoding depends heavily on the characteristics of the input data and the level of noise present in the quantum system, revealing important trade-offs between accuracy and robustness. By developing a fully differentiable training pipeline, the researchers provide practical guidance for selecting the most appropriate encoding method under realistic constraints, ultimately advancing the development of practical quantum machine learning algorithms. Experiments reveal that Angle encoding attains higher accuracy and remains comparatively robust as noise increases when processing aggressively downsampled inputs, specifically at a resolution of 4×4 pixels. At 8×8 resolution, the Hybrid scheme surpasses Angle encoding under moderate noise, suggesting that combining phase and angle injections benefits from increased feature bandwidth., Amplitude-encoded QCNNs, while sparsely represented in downsampled grids, deliver strong performance in both lightweight and full-resolution configurations, exhibiting training dynamics closely resembling classical convergence.

The team trained QCNNs on downsampled binary variants of both MNIST and Fashion-MNIST datasets, employing a data-driven approach to select binary class pairs that maximize separability at extreme compression. Measurements confirm that the performance of each encoding strategy is sensitive to both the input resolution and the strength of depolarizing noise applied to the quantum circuits., The researchers developed a fully differentiable PyTorch pipeline, incorporating a custom autograd bridge and batched parameter-shift gradients, to enable efficient training and evaluation. This pipeline, coupled with a shot scheduling policy, facilitated practical training across a wide range of encoder and noise settings. The results provide practical guidance for selecting QCNN encoders under constraints of resolution, noise strength, and simulation budget, offering valuable insights for advancing near-term quantum machine learning applications.,. Encoding Strategies and Noisy QCNN Performance This work presents a comparative study of three encoding strategies, Angle, Amplitude, and Hybrid, for quantum convolutional neural networks (QCNNs). Researchers developed a fully differentiable training pipeline to evaluate these methods under noisy conditions, using downsampled binary variants of the MNIST and Fashion-MNIST datasets. Importantly, the team highlights that optimal encoding depends on specific constraints, including resolution, noise levels, and available computational resources. From an implementation perspective, the research demonstrates a custom autograd bridge, combined with efficient gradient calculation and shot scheduling, enables large-scale QCNN experiments without relying on approximate gradients. Future research directions include expanding the experimental coverage for Amplitude encoders, increasing QCNN depth and qubit count to explore variational trainability, and applying the framework to additional datasets and real quantum hardware. These efforts will help determine whether the observed trade-offs between Angle and Hybrid encoding persist beyond simulated environments. 👉 More information 🗞 A Comparative Study of Encoding Strategies for Quantum Convolutional Neural Networks 🧠 ArXiv: https://arxiv.org/abs/2512.12512 Tags: Rohail T. As a quantum scientist exploring the frontiers of physics and technology. My work focuses on uncovering how quantum mechanics, computing, and emerging technologies are transforming our understanding of reality. I share research-driven insights that make complex ideas in quantum science clear, engaging, and relevant to the modern world. Latest Posts by Rohail T.: Two-dimensional Quasicrystals Enable Exploration of Unique Magnetic Order and Critical Exponents December 17, 2025 Fluxonium Qubit Achieves Microwave Slow Light and Storage in Single-Atom System December 17, 2025 Quantum Biosensing Achieves 30-Minute Earlier Bacterial Growth Detection December 17, 2025