Multiclass classification using quantum convolutional neural networks with hybrid quantum-classical learning

Abstract

Multiclass classification is of great interest for various applications, for example, it is a common task in computer vision, where one needs to categorize an image into three or more classes. Here we propose a quantum machine learning approach based on quantum convolutional neural networks for solving the multiclass classification problem. The corresponding learning procedure is implemented via TensorFlowQuantum as a hybrid quantum-classical (variational) model, where quantum output results are fed to the softmax activation function with the subsequent minimization of the cross entropy loss via optimizing the parameters of the quantum circuit. Our conceptional improvements here include a new model for a quantum perceptron and an optimized structure of the quantum circuit. We use the proposed approach to solve a 4-class classification problem for the case of the MNIST dataset using eight qubits for data encoding and four ancilla qubits; previous results have been obtained for 3-class classification problems. Our results show that accuracies of our solution are similar to classical convolutional neural networks with comparable numbers of trainable parameters. We expect that our finding provide a new step towards the use of quantum neural networks for solving relevant problems in the NISQ era and beyond.

Quantum machine learning for image classification

“…This potential arises from the unique features of quantum computing, such as superposition and entanglement, which can provide an exponential speedup for specific machine learning tasks [17]. QML algorithms produce probabilistic results, which align well with classification problems [18]. They also operate in an exponentially larger search space, which has the potential to enhance their performance [19][20][21].…”

“…In the context of image classification, QML algorithms can process large datasets of images more efficiently than classical algorithms, leading to faster and more accurate classification [24]. Recent studies have also explored hybrid quantum-classical CNNs and demonstrated the classification [18,[25][26][27][28][29][30] and generation [31][32][33][34] of images.…”

Characterization of variational quantum algorithms using free fermions

“…Originally proposed in the context of quantum chemistry [76] and classical optimization problems [28], such algorithms have since found wide use as a tool for leveraging the current generation of noisy, intermediate scale quantum computers [77] to tackle problems which are hard to solve classically. For that purpose they have been extended to a myriad of domains, including machine learning, [1,9,18,27,38,81], preparation of general condensed matter quantum states [29,41,46,79,80,101,105], finance [25,71], molecular biology and biochemisty [10,74], and linear algebra [11,17,106], and have been implemented in numerous quantum simulation platforms [5,34,48,75,102].…”

A Multi‐Class Quantum Kernel‐Based Classifier

“…Some of the recent work in quantum machine learning has seen the development of multi-class quantum and quantuminspired classifiers that avoid these heuristic strategies. [9][10][11][12][13][14][15][16] Most recently, the quantum-inspired methods [15,16] use techniques from quantum state discrimination for multi-class classification. Other work has seen the development of quantum convolutional neural networks (QCNNs) for multi-class classification.…”

“…Other work has seen the development of quantum convolutional neural networks (QCNNs) for multi-class classification. [9,12] In these methods, a QCNN is trained to reduce the cross entropy loss between its output, interpreted as a probability distribution over all the classes, and the one-hot encodings of the classes.…”