Quantum computing and AI may someday work together
Nespix/Shutterstock
Quantum computers may eventually be able to handle some AI applications that currently require vast amounts of traditional computing power. Such developments would provide a major boost to machine learning and similar artificial intelligence algorithms.
Quantum computers hold the hope that they will eventually be able to complete certain calculations that are not possible with classical computers. Researchers have long debated whether these advantages over traditional computers also extend to tasks involving large amounts of data and the algorithms that learn from it: machine learning, the basis of many AI programs.
Now, Xinyuan Huang and colleagues at quantum computing company Oratomic argue that the answer should be yes. Their mathematical research aims to lay the foundation for a future in which quantum computers provide a widespread boost to AI.
“Machine learning is used not just in science and technology, but in everyday life. In a world where it can be built, [quantum computing] “I feel like this architecture can be applied whenever large datasets are available,” he says.
His team’s research addresses the important question of how data collected in the non-quantum world, such as restaurant reviews or RNA sequencing results, can be fed into quantum computers so that they can leverage the computer’s quantum nature to more efficiently process and learn from the data.
This requires all data to be “overlaid”. This is a mathematical combination that non-quantum machines cannot create. But until now, researchers had considered this task impractical. That’s because it was assumed that all the data in that superposition would need to be stored in a dedicated memory device before being processed by the quantum computer, which would have to be incredibly large, said Caltech team member Hymen Zhao.
Huang and colleagues took a different approach that doesn’t require such memory. This involves inputting data into a quantum computer in small batches without saving everything before starting processing. This is similar to streaming a movie instead of downloading it completely before watching.
They showed that not only does this approach work, but quantum computers can process more data with less memory cost than traditional computers.
In fact, the memory advantage is so great that a quantum computer built with about 300 error-proof building blocks called logical qubits would outperform a classical computer built using every atom in the observable universe, Zhao said.
Although it will likely be many years before a quantum computer with 300 logical qubits can be built, Huang says a computer with 60 logical qubits could probably be built by the end of the decade. At this scale, the researchers’ analysis suggests that some tasks involving processing large datasets and where AI is used already have significant quantum advantages over classical computers.
“Quantum machines are very powerful devices, but first they need to be fed. In this study, we talk about feeding and why loading them is enough. [data] Don’t overfeed the animals, but little by little,” says Adrian Pérez Salinas of the Swiss Federal Institute of Technology Zurich.
Nevertheless, he says many questions still need to be addressed when applying the new research to real devices and real-world data. Many past quantum machine learning algorithms ultimately turned out to be amenable to “inverse quantization.” This is a process by which algorithms are adapted so that they do not require quantum hardware while maintaining good performance. It will also be important to examine how important quantum nature is to this new algorithm, Perez-Salinas says.
Vedran Duniko of Leiden University in the Netherlands said the new research could be suitable for large-scale scientific experiments such as the Large Hadron Collider, where millions of gigabytes of data are continually created, much of which is discarded due to lack of computer memory.
But instead of data centers full of traditional servers, quantum computers will likely only be able to handle some of today’s AI applications and similar types of data processing, he says. “This is not a big part of what GPUs are warming the planet for, but it could still be important,” Dunjko says.
The researchers are now working both to expand the types of algorithms for which their method could be useful and to devise new ways to configure quantum computers, allowing them to process data fast enough with very little memory and in a practical amount of time.
topic:
- artificial intelligence/
- quantum computing
