Commun. Quantum neural networks take a giant leap forward in 'learning'

Now, scientists at EPFL have shown that even simple examples of quantum machine learning models - "quantum neural networks" - can be enough to learn and predict the behavior of quantum systems, bringing us closer to a new era of quantum computing.

Thanks to a pioneering study led by Professor Zoe Holmes and her team at EPFL, we are now closer to achieving quantum reality. Working with researchers at Caltech, Freie Universität Berlin and Los Alamos National Laboratory, they have found a new way to teach quantum computers how to understand and predict the behavior of quantum systems.

The research has been published in Nature Communications.

In the future, computers could demystify quantum mechanics, enabling us to study the behavior of complex materials or simulate the complex dynamics of molecules with unprecedented precision.

The researchers studied quantum neural networks (QNNs), a machine learning model designed to learn and process information using principles inspired by quantum mechanics to mimic the behavior of quantum systems.

Like neural networks used in artificial intelligence, QNNs consist of interconnected nodes or "neurons" that are used to perform computations. The difference is that in a QNN, the neurons operate according to the principles of quantum mechanics, allowing them to process and manipulate quantum information.

"Usually, when we teach a computer something, we need a lot of cases." Holmes said, "But in this study, we show that with just a few simple examples, called 'product states,' computers can learn how quantum systems behave, even when dealing with entangled states (which are more complex and challenging to understand)!"

The multiplicative state used by scientists is a concept in quantum mechanics that describes a specific type of state of a quantum system.

For example, if a quantum system is composed of two electrons, then its product state is formed when the states of each individual electron are considered independently and then combined.

Multiplicative states are often used as a starting point for quantum calculations and measurements because they provide a simpler and more manageable framework for studying and understanding the behavior of quantum systems; and then move on to more complex and entangled states in which the particles are correlated and cannot be described independently.

The researchers demonstrate that by training QNNs with only these simple examples, computers can effectively grasp the complex dynamics of entangled quantum systems.

Applications of Quantum Dynamics Learning

Training results

Holmes explains, "This means [we] may be able to use smaller, simpler computers to learn and understand quantum systems. For example, instead of the noise-containing intermediate-scale (NISQ) computers we may have in the next few years, large and complex computers will be needed. Of course, this could be decades away."

The work also opens up new possibilities for using quantum computers to solve important problems, such as studying complex new materials or simulating the behavior of molecules.

Finally, the method improves the performance of quantum computers by creating shorter, more fault-tolerant programs. "By learning how quantum systems behave, we can simplify the programming of quantum computers, leading to increased efficiency and reliability." Holmes said, "We can make quantum computers better by making their programs shorter and less error-prone."

Reference links:

[1] https://www.nature.com/articles/s41467-023-39381-w#Fig1

[2]https://scienceblog.com/538608/quantum-computers-embrace-quantum-mechanics/

[3] https://phys.org/news/2023-07-quantum-neural-networks-easier.html