For the first time, physicists in the United States have demonstrated a fully functional neural network for image recognition running on an ion-based quantum computer. The study revealed that modern quantum processors are capable of efficiently executing such computations, even in the presence of noise.
According to TASS, citing a publication in the journal Physical Review Letters, the authors of the study believe this experiment confirms the potential for practical applications of existing quantum computers in the field of artificial intelligence. In their view, similar projects previously remained mostly at the level of theoretical discussions.
The neural network was developed by a research team led by Professor Victor Galitski from the University of Maryland. The system is based on a quantum analog of a binary multilayer perceptron, where individual qubits serve as the functional equivalents of neurons.
The researchers arranged the interaction of qubits in a way that mirrors the operational principles of traditional neural networks. This tailored configuration enabled the same algorithm to run on different types of quantum computing devices.
During the experiments, the system performed successfully both on a quantum processor using ytterbium ions, developed at the University of Maryland, and on IBM’s cloud platform featuring superconducting qubits. The neural network utilized 16 qubits for image processing and digit recognition.
The tests showed that the new system outperformed a classical neural network with an equivalent architecture. Moreover, the actual results surpassed the predictions made through simulations. The researchers now plan to investigate how quantum noise enhances the accuracy of image recognition.
About the Author
