Quantum computers as learning machines

Can quantum effects help computers to learn? Image (c) VLADGRIN/shutterstock

Can quantum effects help computers to learn? Image (c) VLADGRIN/shutterstock

A new project on Quantum Optimisation and Machine Learning is now underway. Based at the University of Oxford, it's a joint endeavour between the University, Nokia and Lockheed Martin. The aim of the project is to understand the potential for quantum technology to enhance optimisation and machine learning tasks - these are some of the hardest and most important applications in computer science today.

Machine learning refers to a variety of applications where computers figure out 'for themselves' how to perform data analysis, modelling and inference: tasks range from image and speech recognition through to language translation and even genome analysis. Optimisation involves finding the best solution to a problem from a set of alternatives. Generally these areas are regarded as hard for conventional computers, but they are also extremely important: advances in machine learning and optimisation could greatly increase the range of things that computers can do for us. For example, it may allow computers to be smarter at helping people and companies to manage the ever increasing torrent of information flowing from online systems of all kinds (e.g. smartphones). It is believed that harnessing quantum effects can lead to machines that are fundamentally better at machine learning and optimisation, thus unlocking this potential.

Quantum information processing is a field of research and development that hopes to harness the deepest phenomena in physics in order to create whole new kinds of technology. Various approaches are being taken, all of which are of interest to the QuOpaL project. One particularly interesting approach is adiabatic quantum optimisation (and the closely related phenomenon of quantum annealing). Here, a system is initialised to a simple state and then the conditions are slowly ('adiabatically') changed to reach a complex final state that describes the solution to a computational problem of interest. Many believe that this approach is the best way to start using quantum effects for accelerated machine learning -- whether or not this is true is a key topic of interest to the QuOpaL project!

For more information, including positions available, see the QuOpaL project webpage.

Posted on July 26, 2014 .

Long-lived Qubits Survive for Record Times

Maggie McKee

It may not sound long but, to a quantum physicist, 39 minutes is a veritable eternity. That’s the length of time for which fragile quantum states have now been shown to survive in a solid material at room temperatures in the lab. This record-breaking feat -- carried out by an international collaboration, including Stephanie Simmons at Oxford University -- overcomes a key barrier to building practical quantum computers. Their study is published today in Science.

A quantum bit stored in the heart of phosphorus atom can survive even at room temperature, when the surrounding lattice of silicon atoms form a furiously vibrating 'noisy' environment. Image © 2013 Karl G. Nyman with CC BY license.

A quantum bit stored in the heart of phosphorus atom can survive even at room temperature, when the surrounding lattice of silicon atoms form a furiously vibrating 'noisy' environment. Image © 2013 Karl G. Nyman with CC BY license.

"This opens up the possibility of truly long-term coherent information storage at room temperature," says quantum physicist Mike Thewalt, who performed the test at Simon Fraser University, in Burnaby, British Columbia, Canada, with colleagues.

Quantum computers promise to significantly outperform today’s machines at certain tasks, by exploiting the strange properties of subatomic particles. Conventional computers process data stored as strings of 1s or 0s. But quantum objects are not constrained to the either/or nature of binary bits. Instead, each quantum bit, or “qubit”, can be put into a "superposition" of both 1 and 0 at the same time, enabling them to perform multiple calculations simultaneously. This ability to multitask could allow quantum computers to crack seemingly secure encryption codes, for instance.  "A rather ambitious supercomputer would be able to obtain a result in a million years, whereas a moderate quantum computer would be able to do it in an hour," says Simmons.

The problem with attempts to build these extraordinary number-crunchers, however, is that superposition states are delicate structures that can collapse like a soufflé if nudged by a stray particle, such as an air molecule. To minimize this unwanted process, physicists often cool their qubit systems to almost absolute zero (-273 °C) and manipulate them in a vacuum. But such setups are finicky to maintain and, ultimately, it would be advantageous  for quantum computers to operate robustly at everyday temperatures and pressures.

Now Thewalt, Simmons and colleagues have made a significant step towards this goal of bringing quantum computers into the mainstream. They began with a sliver of  silicon -- around the size of a pencil eraser -- doped with small amounts of other elements, including phosphorus. The quantum information is encoded in the nuclei of the phosphorus atoms. Each nucleus has an intrinsic quantum property called “spin”, which acts like a tiny bar magnet when placed in a magnetic field. Spins can be manipulated to point up (0), down (1), or any angle in between, representing a superposition of the two other states. 

A crucial step to help reduce decoherence was to ionize the phosphorus, removing one electron from each atom with a laser. The reason for this is that phosphorus atoms have five outer electrons, but only four of them bind to the silicon atoms that surround them in a crystal lattice. That leaves the extra electron free to orbit the nucleus and electrically tug on the nuclear spin, disrupting its superposition. Once this electron has been excised, the system is far less likely to decohere.

The team prepared their sample at just 4 °C above absolute zero (-269 °C) and placed it in a magnetic field. Additional magnetic field pulses were used to tilt the direction of the nuclear spin and create the superposition states.  When the sample was held at this cryogenic temperature, the nuclear spins of about  37 per cent of the ions – a typical benchmark to measure quantum coherence – remained in their superposition state for three hours. The same fraction survived for 39 minutes when the temperature of the system was raised to 25 °C.  

The team appears to have set records for coherence within a solid material at any temperature, says Thaddeus Ladd (http://www.thaddeusladd.com/), a quantum physicist at HRL Laboratories, LLC, a research firm in Malibu,   California, who was not involved with the study. "We've managed to engineer a system that seems to have basically no noise," Simmons adds. "They're high-performance qubits."

Though 39 minutes may seem short, it takes only one-hundred-thousandth of a second to flip the nuclear spin of a phosphorus ion – the type of operation that would be used to run quantum calculations. So physicists could perform more than 200 million operations before a qubit's superposition caved in. 

"What is perhaps most important is that this is silicon," notes Ladd. Since silicon already forms the backbone of today’s electronics industry, it is an ideal material to use with an eye to fabricating chips with both conventional and quantum processors, enabling computers in the future to use whichever would be most efficient for a given task, suggests Simmons.

But there is still some work ahead before the team can carry out actual quantum computations. The nuclear spins of the 10 billion or so phosphorus ions used in this experiment were all placed in the same quantum state. To run calculations, however, physicists will need to place different qubits in different states.  "To have them controllably talking to one another – that would address the last big remaining challenge," says Simmons.

This story has been picked up by various media outlets, including the BBC.

Posted on November 15, 2013 .

Strange vibrations: Nanotubes strike a quantum chord

Keeping cool: Edward Laid adjusting the cryogenic core which chills nanotube devices to an operating temperature near absolute-zero, using a Triton from Oxford Instruments.

Keeping cool: Edward Laid adjusting the cryogenic core which chills nanotube devices to an operating temperature near absolute-zero, using a Triton from Oxford Instruments.

Vibrating carbon nanotubes offer one route toward the ultimate computer -- and may even reveal how reality itself emerges from uncertainty

By Kate Becker

Take a one-atom-thick sheet of carbon, roll it into a hollow cylinder, and you have a carbon nanotube: a remarkably strong material so thin that you could fit 100,000 nanotubes across the width of a human hair. Last month, electrical engineers built the first working computer from these tiny cylindrical molecules -- an important step towards creating smaller, faster and lower-powered machines. But materials scientist Edward Laird has a more ambitious aim: He is setting up a new lab at Oxford University to grow these tubes and investigate how they could be used to make a new breed of quantum computers. Plucking them may even reveal how the reality we see around us emerges from a fuzzy and uncertain quantum underworld. 

Laird was first attracted to researching quantum mechanics because of its many bizarre qualities. For instance, before you measure them, quantum particles can exist in a “superposition state”, holding multiple properties at the same time -- an ability that lies at the heart of quantum computers. Traditional computers store and process information in bits, each of which represents either a one or a zero. By contrast, quantum computers operate on quantum bits, or qubits, that can take the value of zero or one, or a superposition of both at once. This allows them to perform many calculations in parallel, running exponentially faster than a conventional computer. “Quantum theory began as a set of rules, like the uncertainty principle, that limit what we can do", explains Laird, "But we've come to understand that by harnessing the quantum world we can achieve what is otherwise practically impossible”.

An electron makes a handy qubit because, before it is measured, its angular momentum, known as “spin”, exists as a superposition of being both “up” (the equivalent of a zero bit) and “down” (the equivalent of a 1). But keeping qubits in superposition is a delicate business. Like a soap bubble that pops as soon as it is touched, superposition breaks down the moment the qubit interacts with the outside environment. For practical quantum computing, physicists must find a way to protect spins for at least several thousand processor cycles. 

Today’s leading spin qubit material, the semiconductor gallium arsenide, can hold a superposition for about 200 microseconds, but only with difficulty, because the atomic nuclei within the compound act like tiny magnets, disturbing the electrons. This is where carbon nanotubes come in: Because the carbon nuclei of the nanotubes are not magnetized, Laird and his colleagues hope that they may be able to maintain a superposition in a nanotube for as long as ten seconds. 

Laird and his colleagues Leo Kouwenhoven and Fei Pei, both of the Delft University of Technology in the Netherlands, demonstrated the first nanotube qubits earlier this year (Laird et al, Nature Nanotechnology 8 565-568 (2013)). Unfortunately, they could only hold their electrons in superposition for about 65 nanoseconds -- less than a billionth of the most optimistic theoretical predictions. One of Laird’s first goals at his new lab is to extend this time, using an ultra-pure form of carbon that has been cooled to within a fraction of a degree above absolute zero and rigorously stripped of stray magnetized nuclei. The photograph shows the core of a dilution fridge, one of the systems that is essential to Laird's work -- these state-of-the-art facilities were funded in part by generous philanthropic donations.

To make a functional quantum computer, you also need a way to control the state of the qubits. Carbon nanotubes, again, offer up a useful solution. Though the carbon nuclei are demagnetized, the nanotube itself carries an internal magnetic field, pointing along the direction of the nanotube, which is felt by electrons moving in the tube. Laird and his colleagues have developed a technique for stamping a bend into the nanotube, causing the magnetic field to change direction and providing a means by which to manipulate the qubits as they move across the kink. 

Carbon nanotubes are more than just tools for building better computers, though. They could help explore one of the deepest mysteries in physics: how the strange laws of quantum mechanics transform into the ordinary rules of physics that we experience on human scales. In the micro domain of atoms, energy is “quantized”. That is, atoms jump from one discrete energy level to another, and can’t land anywhere in between. 

Laird plans to investigate whether nanotubes, which bridge the size gap between the atomic and macro regimes, follow similar rules. One way to check this is by using an electric field to “pluck” nanotubes “like little guitar strings,” he explains. This will determine whether the nanotubes act like macroscopic objects, vibrating with any arbitrary amplitude. If, however, they skip from one amplitude to the next without any crescendo in between, then they are exhibiting quantum behaviour -- and Laird will be a step nearer to understanding how big objects can get, while still obeying quantum rules. Maybe then, he and his colleagues will be able to reveal how the strange notes of quantum mechanics combine to create the familiar melody of the everyday world. 


Posted on October 1, 2013 .

Why you should explode your quantum computer (and wire up the pieces)

Image: An artists depiction of a network of small quantum systems with the overall topology of a torus. (c) Karl Nyman 2013.

Image: An artists depiction of a network of small quantum systems with the overall topology of a torus. (c) Karl Nyman 2013.

In the last couple of years there has been rapid progress toward realising quantum technologies. The qubits (quantum bits) upon which such technologies are founded have been improving, with longer lifetimes and better levels of control. But one thing remains elusive: scaling the few-qubit prototypes in the laboratory up to the many-qubit systems that can tackle real world tasks. 

In response to this challenge, a number of researchers worldwide have been studying the possibility of a network approach: instead of trying to create a single 'monolithic' quantum cpu with many thousands or millions of qubits, instead focus on making small systems with only a few qubits, and wire them up to form the full scale machine. It sounds like an attractive solution, but there has been a problem: the 'wires' that have been demonstrated so far are very error prone, and consequently the operations within each little unit must be almost perfect in order to counteract the errors on the links. Or so it seemed, but now a joint Oxford-Singapore team writing in the journal Nature Communications have described a method by which one can have the best of both worlds: the scalable flexibility of a 'noisy network', yet without any impractical demands on the precision of qubit control. The team, led by Simon Benjamin of Oxford's Quantum and Nanotechnology Theory group (QuNaT), explained their work to Oxford science blogger Pete Wilton.

Posted on July 29, 2013 .