- A big problem we face when we study these quantum systems [without machine learning] is how to deal with this complexity,” Melko told me.
- Read More: Here’s How to Build the First Large-Scale Quantum Computer
The question posed by Melko and other pioneers of the field of quantum machine learning was whether neural nets could perform tasks that are beyond the capacity of algorithms which don’t incorporate machine learning, like modeling the wave function of a multi-particle system—and they didn’t have to wait long for an answer.

- According to a study published last week in Science, two physicists that were not affiliated with Melko created a relatively simple neural network that was able to reconstruct the wave function of a multi-particle system, and it did so better than any previous technique that did not use machine learning.
- This is also a good sign for Melko, who has placed his bets on the benefits of quantum machine learning and created a neural network that was able to identify phase transitions in condensed matter systems.
- Advances in machine learning may very well lead to advances in quantum computing
“We can’t fully mathematically describe the wave function, it’s too complex,” Melko said.

Machine learning could lead to unprecedented advances in quantum physics.

@motherboard: *Intelligent machines are teaching themselves quantum physics*

Last year, Google’s DeepMind AI beat Lee Sedol at Go, a strategy game like chess, but orders of magnitude more complicated. The win was a remarkable step forward for the field of artificial intelligence, but it got Roger Melko, a physicist at the Perimeter Institute for Theoretical Physics, thinking about how neural networks—a type of AI modeled after the human brain—might be used to solve some of the toughest problems in quantum physics. Indeed, intelligent machines may be necessary to solve these problems.

“The thing about quantum physics is it’s highly complex in a very precise mathematical sense. A big problem we face when we study these quantum systems [without machine learning] is how to deal with this complexity,” Melko told me.

“DeepMind winning this game of Go kind of crystallized some of our thinking. Go is a very complex game, but there was a solution that came from machine learning,” he continued. “So we thought, why can’t we employ similar solutions to tackle quantum physics complexity problems?”

As an example, Melko cites his own work, which focuses on condensed matter physics—basically the science of interactions between many quantum particles in various solids or liquids. As Melko wrote in a recent article for Quartz, condensed matter physics “deals with the most complex concept in nature: the quantum wave function of a many-particle system.” The quantum wave function of a particle mathematically describes all of its possible states, or as Melko describes it to me, it is the “infinitely complex…reality of the particle.”

While “infinitely complex” might seem like a bit of an overstatement, according to Melko, just modeling the wave function of a nanometer-scale mote of dust would require a computer whose hard drive contained more magnetic bits than there are atoms in the universe. As for trying to compute the wave functions of several of these dust particles at once with a classical computer? Forget about it.

Read More: Here’s How to Build the First Large-Scale Quantum Computer

The question posed by Melko and other pioneers of the field of quantum machine learning was whether neural nets could perform tasks that are beyond the capacity of algorithms which don’t incorporate machine learning, like modeling the wave function of a multi-particle system—and they didn’t have to wait long for an answer.

According to a study published last week in Science, two physicists that were not affiliated with Melko created a relatively simple neural network that was able to reconstruct the wave function of a multi-particle system, and it did so better than any previous technique that did not use machine learning. As Giuseppe Carleo, a physicist at ETH Zurich and co-author of the study, told New Scientist, “It’s like having a machine learning how to crack quantum mechanics, all by itself.”

Now that Carleo and his colleague have their proof of concept, they hope to develop a more robust neural network that can handle more complex problems. This is also a good sign for Melko, who has placed his bets on the benefits of quantum machine learning and created a neural network that was able to identify phase transitions in condensed matter systems.

“Condensed matter has its own set of benchmark problems and there’s parts of our theories that we don’t understand,” Melko said. “So we applied machine learning to standard problems of condensed matter physics that we might already have solutions for, basically to see how neural networks handle these high levels of complexity that happen in condensed matter.”

As detailed in a paper published Monday in Nature Physics, Melko’s neural network is only a slightly modified version of an AI software used to identify numbers written by humans. Remarkably, this relatively basic machine learning algorithm was nevertheless capable of recognizing different phases of matter in a quantum system, with minimal adjustments. After running the algorithm on some standard condensed matter problems, Melko ratcheted up the complexity of what was being fed to the machine learning algorithm to see how far he could push it before it broke.

Advances in machine learning may very well lead to advances in quantum computing

“We can’t fully mathematically describe the wave function, it’s too complex,” Melko said. “But by studying how algorithms respond to different complexities of condensed matter problems, we’re also studying what makes these algorithms tick. That is what the field of quantum machine learning is, I think: trying to understand whether or not machine learning helps us in some fundamental way on these quantum problems.”

Both Melko and Carleo’s studies bode well for the future of quantum machine learning. As demonstrated by a conference hosted by the Perimeter Institute last year on the subject, there is already quite a bit of interest in how machine learning can be applied to quantum physics—the conference was attended by leading academics in the fields of quantum physics and artificial intelligence, as well as researchers from companies like Google and Intel.

“Before that conference we had never thought about applying machine learning to these many-body quantum physics problems,” Melko said. “But now people are running with the idea of using neural networks to create efficient representations of quantum systems. This field is just starting to take off.”

Moreover, advances in machine learning may very well lead to advances in quantum computing. Although we are still years away from building the first large-scale quantum computer, the computing power of such a device would revolutionize machine learning, which could in turn improve quantum systems.

In the meantime, machine learning could provide insight into problems that could drastically speed up the rate at which quantum technologies are being developed.

“The ultimate goal of this type of machine learning is to help the scientific and industrial fabrication process of a quantum computer,” said Melko. “We’re in the early stages because we don’t have the hardware, but in the future if you have the quantum computer hardware, you could potentially solve some very complex problems with a quantum computer built by an intelligent machine. It’s fascinating: we’re basically bootstrapping ourselves into this world of quantum computing.”

Get six of our favorite Motherboard stories every day by signing up for our newsletter .

Intelligent Machines are Teaching Themselves Quantum Physics