Mathematics of Quantum Computing

A novel image classification framework based on variational quantum algorithms

Shreya Marali (1RV22CS113)

Sahil Naik (1RV22CS169)
R. V. College of Engineering
 INTRODUCTION
Image classification plays a central role in computer vision, but classical CNN models often rely
on global pooling, which can cause loss of key image features. Although pooling reduces model
complexity, it sacrifices fine-grained information crucial for accurate classification.

To address this, the paper proposes a hybrid quantum-classical framework using Variational
Quantum Algorithms (VQAs). This method eliminates global pooling and leverages the
expressive power of quantum circuits while maintaining fewer parameters, aiming to improve
accuracy without overfitting.

Compared to classical models, the proposed approach shows superior performance, especially in
preserving detail. Simulations also reveal that a well-chosen number of quantum measurements
can outperform traditional setups, suggesting practical benefits for quantum-enhanced
classification.
 RELEVANCE
● Improved Classification Accuracy: The proposed framework helps preserve detailed image features,
leading to more accurate classifications, especially in complex datasets.
● Quantum Advantage Exploration: By applying VQAs, the study pushes forward the practical
integration of quantum computing in machine learning tasks.
● Model Efficiency: Eliminating global pooling while reducing parameter count makes the model both
effective and computationally lean.
● Adaptability to Real Devices: Insights into shot noise and measurement optimization bring the method
closer to real-world quantum hardware deployment.
 LITERATURE SURVEY
Sl No. Title Author Summary

1 Exploiting the Quantum Soronzonbold This study evaluates the use of quantum computing (QC) in processing hyperspectral
Advantage for Satellite Otgonbaatar; satellite imagery, focusing on the deployment of parameterized quantum circuit models
Dieter transpiled into the Clifford+T gate set to assess quantum resource demands. It
Image Processing: Review
Kranzlmüller demonstrates that quantum advantage is achievable only when QML models, requiring a
and Assessment, IEEE substantial number of T-gates, outperform classical HPC models in generalization and
transactions on Quantum exhibit weight symmetry-breaking during training.
Engineering 2024 .

2 Quantum–Classical Image Avinash This s presents a quantum circuit–based data augmentation technique that improves
Processing for Scene Chalumuri; CNN image classification accuracy from 68.65% to 76.03% and is compatible with IBM
Raghavendra Quantum hardware. Additionally, it introduces a hybrid quantum–classical CNN
Classification, IEEE Sensors
Kune; S.Kannan; (HQCNN) that reduces training parameters and achieves 85.28% accuracy on the UC
Letters 2024. B. S. Manoj Merced land-use dataset.

3 Variational Quantum NICLAS This study investigates the use of quantum circuit learning (QCL) and the parameter
Algorithms for Differential SCHILLO, shift rule to solve differential equations (DEs) on NISQ-era quantum hardware.
Equations on a Noisy ANDREAS Implemented on IBM's quantum computer, QCL successfully learns functions and solves
Quantum Computer, IEEE STURM a first-order DE, though hardware noise and high errors in derivative estimation limit
transactions on Quantum scalability and derivative order.
Engineering 2025.
 LITERATURE SURVEY
Sl No. Title Author Summary

4 Comparing classical and Giuseppe Corrente, This study compares classical probabilistic and quantum approaches, including a
probabilistic quantum Carlo Vincenzo custom Quantum Turing Machine (QTM), for solving the Hamiltonian Cycle Problem
algorithms for Hamiltonian Stanzione and (HCP). By applying constraints such as bounded-degree graphs and limited node/edge
Vittoria Stanzione counts, the quantum approach yields non-exponential but notable improvements over
Cycle Problem , JQC, 2023.
existing algorithms.

5 Design of an Iterative Model for Lalitha Kumari This work introduces four quantum-enhanced methodologies for biomedical image
Incremental Enhancements in Pappala; Sai Babu processing: Q-SVM-EFE, QGAN-DA, QECA-FD, and QIR-ND. These methods
Veesam; Kannaiah collectively achieved over 95% classification accuracy, SSIM above 0.90 for synthetic
Quantum Image Processing
Chattu; Janapati data, 95% sensitivity in anomaly detection, and PSNR above 40 dB, while significantly
Using Reinforcement Venkata Krishna;
accelerating training, analysis, and reconstruction processes.
Learning-Based Optimizations, Jyostna Devi
IEEE Access 2023. Bodapati; B.
Tarakeswara Rao

6 Quantum-Assisted Shengyao Wu; This article proposes a Quantum Assisted Hierarchical Fuzzy Neural Network
Runze Li; Yanqi (QA-HFNN) that integrates classical deep learning with quantum neural networks
Hierarchical Fuzzy Neural
Song; Sujuan Qin; (QNNs) to learn fuzzy membership functions for improved image classification.
Network for Image Qiaoyan Wen; Fei Implemented in a simulated quantum environment using a single-qubit circuit,
Classification, IEEE Gao
QA-HFNN achieves accuracy comparable to or better than classical models while
Transactions on Fuzzy demonstrating strong robustness and efficient computational performance.
Systems 2025.
 RESEARCH GAP / OPPORTUNITIES

● Pooling-free image classification using quantum circuits remains underexplored.

● Impact of shot noise and measurement count on classification accuracy needs more detailed study.
● Quantum model interpretability and training stability are still open issues.
● Real-device deployment studies for VQA-based models are limited.
 Road To Phase 2
Chosen Direction: Exploring the simulation of existing VQA-based image
classifiers to study how measurement strategies affect performance in noisy
quantum environments.

Progress: Using pre-implemented quantum-classical hybrid models, simulations

were run across different measurement settings to evaluate trade-offs between
noise resilience and prediction accuracy. Early outcomes suggest that moderate
shot counts provide optimal stability without excessive overhead. These
simulations are currently being extended to test adaptability across datasets and
varying circuit depths.
 REFERENCES
1. Cong, I., Choi, S., & Lukin, M. D. (2019). Quantum convolutional neural networks. Nature Physics, 15(12),
1273–1278.
2. Schuld, M., Bocharov, A., Svore, K. M., & Wiebe, N. (2020). Circuit-centric quantum classifiers. Physical
Review A, 101(3), 032308.
3. Cerezo, M., Arrasmith, A., Babbush, R., et al. (2021). Variational quantum algorithms. Nature Reviews Physics,
3, 625–644.
4. Havlíček, V., Córcoles, A. D., Temme, K., et al. (2019). Supervised learning with quantum-enhanced feature
spaces. Nature, 567(7747), 209–212.
5. Biamonte, J., Wittek, P., Pancotti, N., Rebentrost, P., Wiebe, N., & Lloyd, S. (2017). Quantum machine learning.
Nature, 549(7671), 195–202.