Yesterday, March 16, the TASS news outlet, citing the press service of the Moscow Institute of Physics and Technology, reported that Russian scientists managed to create a full-fledged quantum neural network through the use of a whole chain consisting of superconducting qubits. The development is really unique, plus the experts said that in the end, this neural network is able to perform the tasks with incredibly high accuracy.
“We have found a successful quantum chain structure and learning algorithm that allows us to achieve 94% accuracy for standard multi-label classification problems and 90% accuracy for handwritten decimal digit recognition. The accuracy and stability of the algorithm is confirmed by the cross-validation method,” said Alexei Tolstobrov, an employee of the Laboratory of Artificial Quantum Systems at the Moscow Institute of Physics and Technology.
The authors of the project explained that at the moment one of the main tasks for the entire scientific community and participants in the “quantum race” is precisely the creation of quantum artificial intelligence. In this matter, experts mean the use of all kinds of quantum technologies and computers to increase the speed of neural networks and algorithms based on artificial intelligence, since in the end such a symbiosis will lead to the repetition of the properties of biological neural networks by man-made devices. And, apparently, Russian scientists in this matter bypassed their opponents.
To do this, Russian scientists for the first time used transmon qubits created in Russia as one of the components of the neural network, creating a system that is able to recognize and then classify various image options. In the process of forming the latest neural network, Russian physicists used a set of eight qubits. Half of them acted as a neuron of the machine learning system, and the unique properties of transmon qubits provided specialists with the opportunity to use the same qubits when working with each individual layer of the neural network. This is possible both in the process of training the neural network and in its practical operation when performing the assigned tasks.
During testing, this quantum neural network was able to recognize images of healthy and tumor breast tissue, classify different types of wine, and much more – an accuracy of 94% is extremely high. Accordingly, even with a small number of quantum bits, a quantum neural network can perform a complex task.