Quantum computers improve AI predictions

AI models informed by calculations from quantum computers can more accurately predict the behavior of complex physical systems over time than the current best models that use only classical computers, according to a new study led by researchers at University College London.

The survey results were published in a magazine scientific progresscould improve models that predict how liquids and gases move and interact (hydrodynamics) used in fields ranging from climate science to transportation, medicine, and energy generation.

The researchers say the improved performance is related to the quantum device’s ability to hold large amounts of information more efficiently. That’s because instead of bits switching on or off, 1 or 0 like in classical computers, qubits in quantum computers can be 1, 0, or any state in between, and each qubit can influence any of the other qubits. This means that a few qubits can generate a huge number of possible states.

Senior author Professor Peter Coveney, based at UCL Chemistry and the UCL Center for Advanced Research Computing, said: “To make predictions about complex systems, we either run full simulations that can take weeks, and are often too long to be useful. Or we use AI models, which are faster but less reliable on longer timescales.”

“Our quantum-informed AI model means we can provide more accurate predictions faster. Making predictions about fluid flow and turbulence is a fundamental scientific challenge, but there are also many application areas. Our method can be used for climate prediction, modeling blood flow and molecular interactions, or better designing wind farms to produce more energy.”

Quantum computers have the potential to be much more powerful than the fastest conventional computers, but so far their practical utility has been limited.

To make predictions about complex systems, AI models are trained on large amounts of simulated or observed data. In quantum-informed methods, this data is first fed to a quantum computer that learns important statistical patterns, or invariant statistical properties, of the data, i.e., patterns that remain the same over time. These quantum learning patterns are then incorporated into the training of AI models on traditional supercomputers.

Compared to AI models that do not use quantum learning patterns, the quantum-informed method was about one-fifth as accurate at predicting how complex, chaotic systems will behave, was stable over time, was much more efficient, and required hundreds of times less memory.

This efficiency is the result of two quantum properties. One is quantum entanglement, where each qubit can influence other qubits, regardless of distance. Another, superposition, means that a qubit can exist in different classical states at the same time until it is measured. These properties mean that quantum computers with just a few qubits still have enormous computational power.

Lead author Maida is based at the UCL Center for Computational Science. “Our new method appears to demonstrate ‘quantum superiority’ in a practical way, meaning that quantum computers outperform what is possible with classical calculations alone. These discoveries could lead to the development of new classical approaches that achieve even higher accuracy,” Wang said. “The next step is to scale up the method using larger datasets and apply it to real-world situations.” Furthermore, a provable theoretical framework will be proposed. ”

The other lead author, Xiao Xue, based at UCL Advanced Research Computing, said: “This study demonstrates for the first time that quantum computing can be meaningfully integrated with classical machine learning techniques to tackle complex dynamical systems, including fluid dynamics. It is exciting to see this type of ‘quantum-informed’ approach progressing towards practical application.”

The researchers suggested that the ability of quantum computers to compactly capture the underlying physics of such complex systems may be due to the system’s “quantum-like” chaos, where movements in one part of the system affect other parts of the system over large distances (much like entanglement).

They say their method avoids the limitations of current quantum computers (which are very noisy, prone to errors and interference, and require far too many measurements) by using a quantum device at just one step in the process, rather than passing data back and forth between classical and quantum systems.

The study used a 20-qubit IQM quantum computer linked to traditional supercomputing resources at the Leibniz Supercomputing Center in Germany.

To achieve a quantum state, a quantum computer is cooled to -273 degrees Celsius (close to absolute zero and colder than any other temperature in the universe).

The researchers received funding from UCL and the UK Engineering and Physical Sciences Research Council (EPSRC), with support from the IQM Quantum Computer and the Leibniz Supercomputing Center in Munich.