Quantum computers hold the key to achieving what is considered impossible with today’s conventional computing systems. While a fully functional one has yet to be created, quantum simulators — or smaller systems meant to solve specific problems — have already displayed the ability to outperform modern supercomputers at certain tasks.

These quantum structures can run an innumerable amount of operations at ludicrous speeds. This might seem like only a benefit, but Dr. Giuseppe Carleo from the Center for Computational Quantum Physics at the Flatiron Institute in New York explains that quantum computers’ biggest asset is actually a major roadblock.

“Checking that your laptop is functioning correctly is fairly straightforward, doing the same for quantum computers is more complicated,” Carleo tells Inverse. “Every time you run a program on them the output is nondeterministic, which results with many answers for one question. This is what makes a quantum computer so powerful, but it also means that it’s harder to assess if those results are completely random or if they’re correct.”

A new technique feeds experimental measurements of a quantum system to an artificial neural network. The network learns over time and attempts to impersonate the quantum system’s behavior. With enough data, scientists can fully reconstruct the quantum system.
A new technique feeds experimental measurements of a quantum system to an artificial neural network. The network learns over time and attempts to impersonate the quantum system’s behavior. With enough data, scientists can fully reconstruct the quantum system.

But Carleo and a group of international researchers have figured out a way to quickly audit complex quantum systems using artificial intelligence. Their study, which was published in the journal Nature Physics on February 26, provides a technique that will be necessary to show that the quantum computers of the future are actually working.

The way quantum systems store information is what makes them so difficult to verify.

The smallest unit of data in a computer is a bit, which must be a one or a zero. Quantum computing systems use “qubits,” which can represent both one and zero simultaneously. This tiny change enables these computers to tackle an unimaginable amount of tasks. A series of 50 qubits can represent 10,000,000,000,000,000 numbers, this would take up petabytes of space in a traditional computer and would be completely impossible for scientists to go back and check.

Carleo and his colleges used machine learning techniques to essentially check the work of quantum systems, something that isn’t feasible using conventional methods.

Classical bit and Qubit compared

“These machines are able to capture the essence of the quantum system in a very compact way,” said Carleo. “Neural networks understand the relevant features in these extremely complex systems more or less automatically. They are able to grasp this complexity and transform it to understand its fundamental structures.”

This isn’t the first time researchers have used A.I. to do something like this, but Carleo’s work is able to analyze more elaborate systems than the research that preceded it.

Qubits are organized into different shapes to solve various problems. Previous neural nets were only able to audit one dimensional systems, so a straight line of qubits. This study was successfully able to check “two dimensional” and “lattice-shaped” arrays of qubits.

“To characterize more general quantum programs, we need to go beyond this one dimensional structures of qubits,” stated Carleo. “Our technique is a step forward in this direction so that we can tackle arbitrary arraignments of qubits.”

This research goes to show that the creation of a fully functional quantum computer will be wholly reliant on machine learning. Without these kinds of deep learning algorithms no matter how many quantum systems scientists assemble, there would be no way of proving they actually work.

A.I. holds the key to the holy grail of modern-day computing.