Innsbruck physicists have allowed artificial intelligence to plan quantum computing operations

Innsbruck physicists have made a lot of progress in the past in manufacturing quantum computers of ever-increasing capabilities. In order to exploit the promising possibilities that the new type of information processing can offer, computational operations that are not infinitely available must be planned more carefully. Researchers now present a new AI-powered concept in the journal Nature Machine Intelligence.

The team, led by the first author of the work, Jurka Muñoz Gil from the Institute of Theoretical Physics at the University of Innsbruck, compared this approach with several now-popular programs for creating images using artificial intelligence (AI), in which image files are used. “Our new paradigm for programming quantum computers does the same thing, but instead of generating images, it generates quantum circuits based on a description of the quantum process to be carried out,” the scientist was quoted as saying. In the university statement.

Since technologies that follow the sometimes strange rules of quantum mechanics, among other things, are so prone to failure, quantum computers are not easy to set up. This also applies to the order of calculations (quantum gates). The same is true here: what is relatively easy to solve in classical computers that follow the rules of classical physics becomes extremely complex in the quantum world.

Therefore, many scientists are working to integrate artificial intelligence systems into the design of circuits in quantum computers, i.e. gate sequences. The Innsbruck researchers, who include Wittgenstein Prize winners Hans Bregel and Florian Füruter, rely on machine learning “diffusion models,” which are also used in image-generating software. Physicists report that they are now able to show that this approach can be used to avoid problems that may arise when training systems.

“In addition, we show that these propagation models are accurate in their results and also very flexible, allowing circuits to be built with different numbers of qubits (the basic unit of quantum information, note) as well as different types and quantities of quantum gates,” Muñoz-Gil says. The system can also take into account the hardware architecture – that is, the type of communication between qubits – with which it is supposed to work: “Since producing new circuits is very cheap, once you train the model, you can also use it to get new insights into quantum processes.”

service: https://dx.doi.org/10.1038/s42256-024-00831-9