New hybrid method achieves 20% higher accuracy and requires fewer resources, using a quantum computer in the training phase of AI models
New research from University College London (UCL) presents a breakthrough in the field of artificial intelligence: combining quantum computing with machine learning models allows for better prediction of the behavior of complex and chaotic systems, such as the flow of liquids and gases.
The researchers developed a hybrid approach, in which a quantum computer is used at a central stage in the training process of the model. Unlike classical computers that use bits (0 or 1), quantum computers use qubits that can be in multiple states simultaneously, as well as influence each other through quantum entanglement. These properties make it possible to represent a wide range of physical states even with a relatively small number of qubits.
Instead of running long, complex simulations, or relying on AI models that can lose accuracy over time, the researchers used a quantum computer to identify stable statistical patterns within the data. These patterns, called invariant features, were then used to train an AI model on a classical supercomputer.
The results were impressive: the combined model was about 20% more accurate than standard models, and remained stable over time even when predicting chaotic systems. In addition, the method required much less memory – up to hundreds of times less – making it particularly efficient for large and complex simulations.
Problems in the field of fluid dynamics
The research focused specifically on problems in fluid dynamics, a central field in physics and engineering. An accurate understanding of the flow of liquids and gases is essential for predicting weather and climate, designing transportation systems, understanding blood flow in the human body, and even designing more efficient wind turbines.
One of the important implications of the study is the demonstration of what is known as a “quantum advantage” – a situation in which quantum computing provides a better result than that achievable by classical means alone. The researchers note that this is one of the first cases in which such an advantage is manifested in a practical application, and not just in theory.
However, quantum computers are still in their early stages of development, and suffer from noise, errors, and interference. To address these limitations, the researchers have integrated the quantum computer into just one step of the process, rather than passing data back and forth between quantum and classical systems.
The study was conducted using a 20-qubit quantum computer, operating at temperatures close to absolute zero (minus 273 degrees Celsius), and connected to a supercomputer at a computing center in Germany. The researchers now plan to expand the method to more complex systems and test its application in real-world scenarios.
In short, the combination of quantum computing and artificial intelligence may mark a new direction in computational science: more accurate, faster, and more resource-efficient models that can improve predictive capabilities in many fields – from climate to medicine.
For the scientific article: DOI: 10.1126/sciadv.aec5049
More of the topic in Hayadan:
