The dream of useful quantum computing may have just come one step closer.
Australian researchers are combining two of the hottest topics in science: quantum computing and machine learning. Specifically, they’ve succeeded in training an algorithm to predict the evolving state of a simple quantum computer. Such an understanding allows real time stabilization of the system, much as tightrope walker uses a pole for balance, according to a paper published Monday in Nature Communications. That would be a big deal for everyone – from Silicon Valley to Washington, D.C.
Quantum computing extends the familiar concept of the bit to propose the "qubit." While we usually etch transistors in silicon, the quantum analog could be a single particle such as a photon or electron. Like the transistor, this particle is able to exist in two states that correspond to 0 or 1. The difference is, the world at the quantum level looks nothing like ours. In addition to being 0 or 1, the particle can occupy a state not purely 0 or 1 but in some sense a mixture of the two. For this reason, a qubit can be much more flexible than a regular bit.
Exploiting this probabilistic messiness is the key to quantum computing.
The mathematical behavior resists simple characterization, but the general idea is that they could take advantage of a phenomenon called interference to analyze many solutions to a problem simultaneously. In the end, more likely solutions would be amplified and less likely solutions eliminated by competing qubits, much like how ocean waves can combine to make superwaves, or cancel out entirely.
This simultaneous solution testing capability makes quantum computers theoretically useful for solving certain types of problems that would usually require a brute force approach, such as factoring large numbers and encryption. However, each problem requires a specialized method, so chances we’ll someday be checking Facebook and playing games on a quantum computer are slim.
This tantalizing dream of super-fast quantum computers not bound by the standard laws of physics has hovered on the horizon for decades, but progress is slow. IBM built a functional five-qubit system, and the record belongs to USC/Lockheed-Martin’s reportedly 1098 qubit D-Wave 2X system, although the topic is so complicated that no one can say for sure if it’s working or not.
What makes it so tricky?
Quantum computing depends on its qubits doing multiple things at once, for example spinning clockwise and counterclockwise at the same time, and interfering with other qubits in a useful way. Such behavior is so rare at our level of reality as to be unimaginable, and recreating it on demand requires an exacting environment, isolated from the destabilizing influence of the outside world. The D-Wave system, for example, operates at two-one-hundredths of a degree Celsius above absolute zero.
As a rough analogy, you could imagine the activities of the qubits are like a tightrope walker at risk of being knocked off balance at any moment by a gust of wind or a lobbed tomato. To protect the walker, we can take defensive measures to block out external influences, say by erecting a glass barrier around them.
The quantum analog of falling off the tightrope is a process called "decoherence," which describes what happens when a system starts to act classically. That’s no good for a quantum computer, which depends on "coherence" for its quantum magic.
To make matters worse, in the quantum world, if we look at the tightrope walker, they fall. "To build a quantum computer," explains University of Toronto physics professor Aephraim Steinberg, "you need to be sure no information leaks out that could possibly tell which one it was,” a "0" or a "1."
In addition to isolation, the Australian team, lead by quantum physicist Michael Biercuk has made progress on a more active form of qubit aid called quantum error correction. Instead of just protecting the tightrope walker, they’re actively helping. Whenever a qubit is about to decohere, they give a stabilizing nudge with a laser or adjust the frequency of oscillation, which would be something like tweaking the angle of the tightrope or having the walker speed up, or slow down, according to Daniel Lidar, professor of electrical engineering at the University of Southern California.
Without such error correction, "quantum computing would have been dead in the water 20 years ago," Dr. Steinberg tells The Christian Science Monitor in an email, but the novel aspect of the paper is how Dr. Biercuk’s team knew what corrections to make. Remember, looking at the tightrope makes the walker fall, so we have to help while blindfolded. As Steinberg puts it, you need a clever scheme to "measure whether an error occurred, and which one, without measuring what state the qubit is actually in."
Biercuk realized that if his team could predict how the qubits would decohere, they could apply the necessary corrections in real time and keep the balancing act going. But how do you guess what a chaotic system you can’t look at is going to do in the future?
Enter machine learning
"We used algorithms which have broad applications in many fields of science and engineering, and are already widely used," Biercuk tells the Monitor in an email. "Much of the power of our finding is that existing machine learning techniques now have a role to play in building quantum tech."
Based on past data of a qubit’s behavior, the team’s algorithm was able to train itself to predict how the system would evolve in the future. The processes governing this evolution are largely random, but there are some patterns the machine is able to detect. "The random behavior we can correct contains within it what are known as "correlations" in time – such processes change in such a way as to exhibit memory of the past state of the system. It is this correlation which we learn and exploit," Biercuk explains.
First, the team trained the program using data from repeated observations of the qubits, finding out whether the tightrope walker was right handed or left handed, if the wind blows primarily from the east, or west. Of course, during observation the quantum computer is useless. To apply what the algorithm had learned, they used a technique called "multiplexing."
For a time, the trained algorithm watched the computer run, absorbing more transient aspects of the system. Was the tightrope walker sleepy that day? Was the room breezy? Then, they closed the box and immediately let the computer operate in the isolation it needs to perform useful calculations. The algorithm forecasted what would likely be happening inside the box, and the system could apply the appropriate corrections, which reportedly led to a significant improvement over previous methods.
While Best Buy may not be stocking its shelves with code-busting quantum supercomputers anytime soon, Biercuk’s method is a new approach with the potential to move the field forward. "This technique joins and nicely complements the existing arsenal of quantum error correction techniques and will undoubtedly find wide use," says Dr. Lidar, who was not involved in this research. "The marriage of machine learning and quantum error correction may prove to be an important step towards the realization of scalable quantum computing."
[Editor's note: This story has been updated to correct which quantum system holds the record for the most qubits. The current record holder is the D-Wave 2X at the USC-Lockheed Martin Quantum Computing Center.]