MISiS, RCC and MSU: Quantum neural network for image recognition

Main article:

• Neural networks (neural networks)
• Computer vision (machine vision)
• Machine Learning

2022: Method of classification of photographs based on quantum convolutional neural network

Russian physicists of the Laboratory of Quantum Information Technologies of the University of MISIS, the Russian Quantum Center and Moscow State University named after M.V. Lomonosov first presented a method of classification of photographs with high accuracy for 4 classes of images, based on the architecture of the quantum convolutional neural network (QCNN). Representatives of NUST MISIS reported this to TAdviser on November 24, 2022.

To do this, scientists have improved the structure of the quantum circuit and the quantum model of the perceptron - a mathematical or computer model of the perception of information by the brain in the form of some logic circuit with transitions, associative and reactive elements, which is an elementary block of the neural network. Scientists tested the proposed classifier on various samples of four images of handwritten numbers or photographs of clothes and shoes.

"For the first time, we have implemented a proposed approach to solve the problem of classifying 4 classes of images - handwritten numbers and items of clothing, using eight qubits to encode data and four auxiliary qubits. The corresponding machine learning procedure was implemented in the form of a hybrid quantum-classical (variation) model. This approach can be implemented both on emulators and on real quantum processors. Quantum machine learning is one of the most interesting areas of use for quantum computers, "explained Alexey Fedorov, head of the laboratory of quantum information technologies at NUST MISIS and RCC.

Scientists tested the proposed classifier on various samples of four images of handwritten numbers or photographs of clothes and shoes

Recently, neural networks have been actively used to solve a wide range of computing problems. Meanwhile, the power of classic computers is no longer growing - according to experts, this means that a new approach to training neural networks is needed to develop machine learning.

Quantum processors, which in the future can manipulate huge amounts of data and surpass classical computers in certain tasks, will allow the implementation of quantum machine learning. When machine learning switches to quantum computers, some of the processes can accelerate several times, and another part - millions, respectively, quantum neural networks will be more efficient and more efficient than ordinary ones, scientists say.

Machine learning methods are already actively used in research in the field of quantum computing, for example, in solving the problem of image classification, which is central to the creation of computer vision.

"Let's say you have several images and you need to somehow classify them, that is, in simple language, the computer should look at the image and assign one of the labels to it. The image itself, of course, the computer does not see, it sees only a huge grid of numbers. In this case, a machine learning method will help to classify images, which uses a large database, learning from them and finding some patterns in the images submitted to the input. We solve this problem using quantum machine learning, which is based on quantum convolutional neural networks, and we see the potential for the development of this approach, "said Alyona Mastyukova, junior researcher at the Laboratory of Quantum Information Technologies at NITU MISIS and RCC.

Quantum convolutional neural networks (QCNNs) are a series of convolutional layers or sequences of quantum operations alternating with combining layers that together reduce the size of stored information while preserving important set functions. data

The obtained results show that the high accuracy of the solution of the method proposed by Russian scientists is similar to the accuracy of classical convolutional neural networks with a comparable number of trained parameters.

Scientists plan to make further optimization of the perceptron more effective so that classification problems are solved significantly faster than by classical methods.

The study was carried out as part of the strategic direction "Quantum Internet" of the Priority 2030 Program, a grant from the Russian Science Foundation and the Roadmap for the Development of Quantum Computing.

An article on research in quantum machine learning is published in the journal Frontiers in Physics.