The researchers at EPFL demonstrate that a quantum machine-learning model known as “quantum neural networks” can learn and predict the behavior of quantum systems from a small number of straightforward examples, bringing us one step closer to a new era of quantum computing.
Imagine a world in which computers are able to decipher the mysteries of quantum mechanics, allowing us to study the behavior of intricate materials and accurately simulate the intricate dynamics of molecules.
We are now a step closer to making that a reality as a result of a ground-breaking study conducted by Professor Zoe Holmes and her EPFL team. They have discovered a novel method for teaching a quantum computer how to comprehend and anticipate the behavior of quantum systems. They did this in collaboration with specialists from Los Alamos National Laboratory, the Free University of Berlin, and Caltech. Nature Communications has published the study.
“When we teach a computer something, we usually require a lot of instances. However, in this paper, we show that by using just a few simple instances known as ‘product states,’ a computer can learn how a quantum system operates even when dealing with entangled states, which are more intricate and difficult to understand.”
Professor Zoe Holmes and her team at EPFL,
Quantum neural networks (QNNs) The researchers worked on “quantum neural networks” (QNNs), a type of machine-learning model that uses quantum mechanics-inspired principles to learn and process information and mimic the behavior of quantum systems.
Very much like the brain networks utilized in man-made consciousness, QNNs are made of interconnected hubs, or “neurons,” that perform estimations. The neurons in QNNs are able to handle and manipulate quantum information because they operate in accordance with quantum mechanics.
Holmes asserts, “Normally, we need a lot of examples when we teach a computer something.” However, this study demonstrates that even when dealing with entangled states, which are more complicated and difficult to comprehend, the computer can learn how a quantum system behaves by using just a few straightforward examples known as “product states.”
Product states The term “product states” that the researchers employed is derived from a concept in quantum mechanics that describes the particular kind of state that exists within a quantum system. For instance, the product state of a quantum system with two electrons is formed when the states of each electron are taken into account separately and then combined.
Before moving on to more complicated and entangled states, in which the particles are correlated and cannot be described independently, product states are frequently used as a starting point for quantum computations and measurements because they provide a framework that is simpler and more manageable for studying and understanding the behavior of quantum systems.
Better quantum computers are on the way. The researchers demonstrated that computers can effectively comprehend the intricate dynamics of entangled quantum systems by training QNNs with just a few of these straightforward examples.
Holmes makes sense of this: “This implies that [we] could possibly find out about and comprehend quantum frameworks utilizing more modest, easier PCs, similar to the close term delegate scale [NISQ] PCs we’re probably going to have before long, rather than requiring enormous and complex ones, which might be many years away.”
In addition, the research provides new avenues for the application of quantum computers to important problems like simulating the behavior of molecules or studying novel, complex materials.
Lastly, the method makes it possible to write shorter, more error-resistant programs, which boosts quantum computers’ performance. By figuring out how quantum frameworks act, we can smooth out the programming of quantum PCs, prompting further development of effectiveness and unwavering quality. ” “By making their programs shorter and less prone to errors, we can make quantum computers even better,” says Holmes.
More information: Matthias C. Caro et al, Out-of-distribution generalization for learning quantum dynamics, Nature Communications (2023). DOI: 10.1038/s41467-023-39381-w
