Artificial Intelligence learns faster with quantum technology
An international collaboration, including Leiden physicist and computer scientist Vedran Dunjko, showed that quantum technology can speed-up the learning process of artificial intelligence (AI). To prove this, the physicists and computer scientists used a quantum processor for single photons. Their work contributes to the advancement of quantum AI for future applications, such as in the quantum internet. The results have been published in Nature.
The race for smarter machines
Computers beating humans in go, recognising human voices, or helping in finding optimal medical treatments: those are only a few astonishing examples of what the field of artificial intelligence has produced in the past years. The ongoing race for smarter machines has led to the questions of how and with what means improvements can be achieved: for example: how can we make these computers learn faster? In parallel, huge recent progress in quantum technologies has confirmed the power of quantum physics, not only for its often peculiar and puzzling theories, but also for real-life applications. Hence, the idea of merging the two fields: on one hand, artificial intelligence with its autonomous machines; on the other hand, quantum physics with its powerful algorithms.
Scientists already showed several fascinating results, such as AI systems deciding faster on their next move, or assisting in the design of new quantum experiments. Yet, these demonstrations were still incapable of learning faster, a key feature to create increasingly helpful autonomous machines. Think of automated driving or safer automated factories. Vedran Dunjko of the applied Quantum algorithms (aQa) initiative at Leiden University’s Faculty of Science explains: ‘In this context, learning faster means “by using fewer interactions with the environment in which we learn”: like a child who learns with fewer tips from the teacher.’
Processor with light particles
The team of physicists and computer scientists from Leiden University, the University of Innsbruck, the Austrian Academy of Sciences, the Massachusetts Institute of Technology and the German Aerospace Center is the first to successfully prove a speed-up in the computer’s learning time. The team took single photons, the particles of light, and coupled them into an integrated photonic quantum processor. The processor served as the core of the learning machine and was also used for the implementation of the learning tasks. ‘The experiment showed that quantum physics significantly reduce the learning time,’ says Valeria Saggio from the University of Vienna, first author of the publication. ‘We are just at the beginning to understand the possibilities of quantum artificial intelligence,’ adds the Austrian Philip Walther, leader of the collaboration.
In its essence, the experiment can be understood by imagining a computer considering a next move when, for instance, learning to playing chess by playing a virtual opponent. The computer learns by obtaining a reward when making the correct move. Now, if the computer is placed in our usual classical world to learn how to play, it will have to test different moves in different games sequentially, and only one will be rewarded as best. In contrast, when the computer exploits quantum technology, the bizarre aspects of quantum physics come into play. The computer can now make use of one of its most famous and peculiar features: the so-called superposition principle. This can be intuitively understood by imagining the computer to try out many moves at the same time, as it were. This is therefore is far more efficient.
Vedran Dunjko made a great contribution to the theoretical foundations of the method. ‘Quantum methods using light particles are particularly well-suited as building blocks for intelligent algorithms in the quantum internet of the future,’ he says. The first steps of this quantum internet are taken in the Netherlands, with a big role for Leiden and Delft. Dunjko: ‘In a quantum internet we can exchange quantum information, use superpositions to ensure computers communicate faster and more securely; it is also all done with light particles, compatible to the principles of our learning experiment.’
Experimental quantum speed-up in reinforcement learning agents", V. Saggio, B. Asenbeck, A. Hamann, T. Strömberg, P. Schiansky, V. Dunjko, N. Friis, N. C. Harris, M. Hochberg, D. Englund, S. Wölk, H. J. Briegel, and P. Walther, Nature, 10 March 2021; DOI: 10.1038/s41586-021-03242-7
Coverpicture: Artistic impression of the quantum learning concept. © Rolando Barry, University of Vienna.
What’s a quantum computer?
For a conventional computer, the smallest unit of information is the bit, which has a value of either 0 or 1. Physically, a bit can take the form of a little electronic switch that is either on or off. Any modern microprocessor contains millions of these. In contrast to the conventional computer, the quantum computer computes uses the qubit, which has the value of both 0 and 1 in a variable mix, using the fundamental quantum principle of superposition. A qubit is something that cannot be comprehended using one’s common sense. It is the product of the almost magical quantum world, and at the core of all quantum computational advantages.