.png){kind=icon}
.png){kind=icon}
Return
Modified transfer learning based variational quantum circuit for MNIST image classification
.png){kind=icon}
OA
AI
DOI：10.1080/09540091.2026.2624939.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Handwritten digit recognition (HWR) is a technology designed to enable mobile devices and computers to interpret handwritten inputs, encompassing both online and offline sources, such as images, scanned documents and digitised content. Various sectors, including insurance, banking, retail, healthcare and logistics, have increasingly adopted HWR. Despite the use of methods such as optical characteristic recognition (OCR), connectionist temporal classification (CTC) and machine learning (ML), their accuracy is compromised by interpretational limitations and significant error probabilities, leading to unreliable results. In order to address these shortcomings, this study advocates the collaborative use of deep ResNet and Inception models, combined with a variational quantum circuit (VQC) for HWR. The investigation utilises the MNIST dataset, which comprises 60,000 training set images and 10,000 testing set images of handwritten digits, each with a size of 28 × 28 pixels. Two selected samples, namely, Sample-1 (1, 0) and Sample-2 (3, 6), are drawn from this dataset. The hybrid approach of ResNet-Inception, which integrates a quantum encoding layer and VQC, is designed to enhance computational precision and digit recognition. ResNet improves accuracy through residual blocks with skip connections that mitigate vanishing and exploding gradients, while VQC enhances efficiency for high-velocity, large-scale datasets. The framework is evaluated using accuracy, precision, recall, and F1-score.
Keywords:
Image classification
deep learning
quantum computing
transfer learning
variational quantum circuits
Journal
IF:
3.4
Papers:
843
Citations:
1.5K