Future and emerging technologies: Quantum computing

Event report

[Read more session reports from WSIS Forum 2017]

The moderator, Mr Catalin Marinescu (Head of Corporate Strategy Division, ITU) opened the session highlighting quantum computing as one of the key trends in innovation in ICTs and emphasising the expectation that it will be available to create value for customers and society within ten years.

Dr Colin P. Williams (Director of Strategy & Business Development at D-Wave Systems Inc.) started his keynote speech by defining a quantum computer as one that harnesses quantum physical effects that are not available to conventional computers. He mentioned three:

• Superposition:  unlike conventional computers, in which a bit assumes binary states of 0 or 1, a qubit in a quantum computer may assume partially 0 and 1 at the same time.
• Entanglement:  if an operation is performed on one subset of the qubits, it changes the state of other qubits which were not directly touched, creating possibilities for the rise of algorithms that function in whole different ways.
• Tunnelling: a property which allows qubits to cross barriers which are unsurmountable in classic processing.

According to Williams, there are many ways to combine quantum effects to build a computer. He mentioned three of these also:

• Gate model: based on an analogy with binary arrangements, but which is too difficult to scale, since it needs massive qubit overhead for error correction.
• Topological model: similar to a gate model that allows to compute without error crash, but which is only theoretical and needs a quasi-particle whose existence and robustness are disputed.
• Annealing: harnesses nature’s ability to find low-energy configurations via tunnelling, which is resilient to noise and handles a wide range of important problems.

His company (D-Wave Systems Inc) works on a non-universal model of a quantum computer, which is adequate to provide solutions for most of the available market and thus justifies a business model. Williams pointed out that a quantum computer can be programmed through calls from regular programming languages to the quantum computer without the need to know physics. He mentioned experiments that showed that an annealing-based quantum computer can be one hundred million times faster than conventional machines for a particular problem, and as a consequence, classic algorithms which rely on existing architectures may become ineffective.

Concerning the applications, Williams mentioned that the enhanced processing power of quantum computing has allowed the company to successfully factor numbers, in a sign that it might be used to break cryptographic keys. He argued, however, that even if that happens, the reliability of encryption can be restored when cryptography itself is implemented using quantum tools, making it even harder to break. In his view, the technology can boost artificial intelligence (AI) techniques with quantum machine learning using the quantum resources to accelerate bottlenecks in solving AI problems. On top of that, he said, quantum computing dissipates a fraction of the energy of conventional computing, and for applications like machine learning, a flat power budget is a huge competitive advantage.

Williams said that he believed quantum computing will not necessarily replace conventional computing, but that a combination of both kinds of processing in a hybrid use will often be more efficient. He mentioned a study which consisted in identifying binary classifiers for Google’s car in which having learned and identified the optimal classifiers using faster and better quantum techniques, the descriptors were deployed back in devices that compute traditionally.

He concluded that quantum computers can make artificial intelligence models feasible, increasing the efficiency of unsupervised learning – machine learning that uses data that has not been humanly labeled. Williams emphasized that the potential of unsupervised learning is particularly important since most of the data in the world is not labeled. Quantum computing may be decisive in this application because unsupervised learning deploy probabilistic models based on sampling, and quantum computers are fast native samplers.

by Cláudio Lucena