Comparing concepts of quantum and classical neural network models for image classification task

TL;DR

This study assesses whether a hybrid quantum-classical neural network can outperform a similarly parameterized classical network on MNIST digit classification when input data are encoded into quantum representations. It employs two parallel PQCs fed by partitioned 4×4 image fragments, with a classical merge and softmax output, and compares performance against a conventional MLP under a controlled 10-epoch training regime. Results show the quantum model converges faster and achieves higher test and balanced accuracies, despite the substantial computational cost of simulating quantum circuits. The work demonstrates the potential of quantum data representations for image classification and motivates further research into encoding strategies and scalable quantum hardware implementations.

Abstract

While quantum architectures are still under development, when available, they will only be able to process quantum data when machine learning algorithms can only process numerical data. Therefore, in the issues of classification or regression, it is necessary to simulate and study quantum systems that will transfer the numerical input data to a quantum form and enable quantum computers to use the available methods of machine learning. This material includes the results of experiments on training and performance of a hybrid quantum-classical neural network developed for the problem of classification of handwritten digits from the MNIST data set. The comparative results of two models: classical and quantum neural networks of a similar number of training parameters, indicate that the quantum network, although its simulation is time-consuming, overcomes the classical network (it has better convergence and achieves higher training and testing accuracy).

Figures (6)

• Figure 1: Graphical description of quantum data processing basics
• Figure 2: Quantum encoding of example image (a) with $\text{X}$ gates (b) broughton2020 and $\text{R}_\text{x}$ gates (c) potempa2021
• Figure 3: Layout of a simplified version of PQC used for experiments, where $(-2, -2)$ qubit is the readout qubit and the other qubits are used for image pixels representation potempa2021.
• Figure 4: Overviews of the classical (a) and quantum neural network (b) model architectures
• Figure 5: Changes of loss function (\ref{['eq:cce']}) for classical and quantum models in ten epochs. Transparent lines relates to single experiment and solid lines are mean results.