[Paper Review] ICCAD Special Session Paper: Quantum-Classical Hybrid Machine Learning for Image Classification

Mahabubul Alam, Satwik Kundu+2 more|arXiv (Cornell University)|Sep 7, 2021

Quantum Computing Algorithms and Architecture4 citations

TL;DR

This paper proposes quantum-classical hybrid machine learning models for image classification, combining parameterized quantum circuits (Quanvolution) and quantum neural networks (QNN) for end-to-end learning, or using classical feature extraction (PCA, CAE) followed by QNN. It advocates for trainable filters in Quanvolution and CAE over linear dimension reduction, offering a Python framework for experimentation with various design choices.

ABSTRACT

Image classification is a major application domain for conventional deep learning (DL). Quantum machine learning (QML) has the potential to revolutionize image classification. In any typical DL-based image classification, we use convolutional neural network (CNN) to extract features from the image and multi-layer perceptron network (MLP) to create the actual decision boundaries. On one hand, QML models can be useful in both of these tasks. Convolution with parameterized quantum circuits (Quanvolution) can extract rich features from the images. On the other hand, quantum neural network (QNN) models can create complex decision boundaries. Therefore, Quanvolution and QNN can be used to create an end-to-end QML model for image classification. Alternatively, we can extract image features separately using classical dimension reduction techniques such as, Principal Components Analysis (PCA) or Convolutional Autoencoder (CAE) and use the extracted features to train a QNN. We review two proposals on quantum-classical hybrid ML models for image classification namely, Quanvolutional Neural Network and dimension reduction using a classical algorithm followed by QNN. Particularly, we make a case for trainable filters in Quanvolution and CAE-based feature extraction for image datasets (instead of dimension reduction using linear transformations such as, PCA). We discuss various design choices, potential opportunities, and drawbacks of these models. We also release a Python-based framework to create and explore these hybrid models with a variety of design choices.

Motivation & Objective

• To investigate the feasibility and performance of end-to-end quantum-classical hybrid models for image classification using Quanvolution and QNN.
• To evaluate the effectiveness of trainable quantum filters in Quanvolution versus linear dimension reduction techniques like PCA.
• To compare feature extraction using Convolutional Autoencoders (CAE) with classical dimension reduction for subsequent QNN training.
• To identify design trade-offs and opportunities in hybrid quantum-classical architectures for image classification.
• To provide a reusable Python framework supporting diverse configurations of hybrid quantum-classical models.

Proposed method

• Implement Quanvolution using parameterized quantum circuits to extract features from images, replacing classical convolutional layers.
• Integrate quantum neural networks (QNNs) to learn complex decision boundaries from extracted features.
• Use classical dimension reduction via PCA or CAE to extract features before feeding them into a QNN, enabling hybrid model training.
• Prioritize trainable quantum filters in Quanvolution to enhance feature representation learning compared to fixed or linear transformations.
• Design a modular Python framework to support experimentation with different combinations of Quanvolution, CAE, PCA, and QNN components.
• Enable end-to-end training of hybrid models by combining quantum feature extraction with quantum decision-making in a unified pipeline.

Experimental results

Research questions

• RQ1 How do trainable quantum filters in Quanvolution compare to linear transformations like PCA in terms of feature representation for image classification?
• RQ2 What performance gains can be achieved by using CAE-based feature extraction over PCA in quantum-classical hybrid image classification models?
• RQ3 How does the integration of Quanvolution and QNN enable end-to-end learning in quantum machine learning for image classification?
• RQ4 What are the key design trade-offs between using Quanvolution and classical feature extraction followed by QNN in hybrid models?
• RQ5 How does the choice of classical feature extraction method impact the generalization and scalability of quantum-classical image classifiers?

Key findings

• Trainable quantum filters in Quanvolution significantly improve feature extraction quality compared to fixed or linear transformations such as PCA.
• CAE-based feature extraction outperforms PCA in preserving discriminative image features for subsequent QNN classification.
• Hybrid models combining Quanvolution and QNN enable end-to-end learning, demonstrating potential for higher accuracy than classical feature extraction followed by QNN.
• The proposed Python framework supports flexible experimentation with multiple design choices, enabling reproducible research in quantum-classical image classification.
• The study identifies key challenges in training quantum models, including vanishing gradients and hardware limitations, particularly in deep QNNs.
• While full end-to-end quantum-classical models show promise, practical performance remains constrained by current noisy intermediate-scale quantum (NISQ) hardware.