sommai - Fotolia

Quantum machine learning may come knocking on analytics' door

Quantum mechanics was discovered 100 years ago. Now, as technology based on it emerges from labs, IT vendors are setting their sights on quantum computing and analytics applications.

Quantum computing emerged from the lab in the 1990s when researchers at the National Institute of Standards and Technology used trapped ions to create the first quantum logic gates. Then, quantum computing went back in the lab, re-emerging on occasion to report fitful steps of progress.

These days, the drumroll of quantum computing is more pronounced -- and a newly emerging notion of quantum machine learning may amplify it further yet.

In September, Microsoft discussed a new programming language integrated with Visual Studio for work on quantum simulators and quantum computers. This month, the company released a preview of what it calls Q#, as part of a free software development kit (SDK).

In October, Intel delivered a superconducting prototype chip to QuTech, a Netherlands-based research center that is its partner in the quantum pursuit, for use in testing quantum concepts.

And in November, IBM vowed to open up its first superconducting IBM Q systems to clients by the end of 2017. Earlier, IBM had introduced an SDK for quantum computing programmers.

All of this isn't to say that quantum computing will soon move out of the realm of research and development. But it may be time for a larger number of IT professionals, including data pros, to give some thought to the topic.

It came out of the lab

The budding potential of quantum machine learning may be borne out by deals like one that recently saw consulting company Accenture form an alliance with startup 1QB Information Technologies Inc. to begin to bring quantum computing analytics to businesses. Earlier this year, Accenture and 1QBit, as the startup calls itself, began work with biotech giant Biogen to develop quantum-oriented tools for drug discovery.

Accenture's experience with analytics and systems integration can help 1QBit fulfill its mission to connect research-oriented quantum work with real-world applications, according to Kausar Samli, COO at the Vancouver, B.C.-based quantum computing platform and services company.

"We hope to come to better understand and build approaches for doing analytics with quantum practices," Samli said. "Accenture has a strong analytics background, and this allows us to dive into the area together."

Application areas of interest are likely to include scheduling, optimization and logistics, as well as social network analysis, chemicals and financial engineering, he said.

Did somebody say 'quantum leap'?

Oil and gas explorers and materials researchers -- not to mention cryptologists -- have long used supercomputers to do tough analytics. But their immense data sets can, at times, stymie even the greatest supercomputers. These folks are interested in new means of computing because they promise a massive leap in capabilities.

Quantum computer potential is based on the promise of quantum mechanics. This was a tricky bit of physics discovered about 100 years ago. Rather than being based, like conventional computers, on bits that hold one of two states, quantum computers are based on qubits, or quantum bits, that can hold up to two states at the same time.

That means supporting states of not just 0 or 1, but, also states of 0 and/or 1. The nuance is slight, but, by exploiting quantum mechanical effects at an atomic scale, the qubit greatly boosts raw computing power. Theoretically, the quantum architecture goes far beyond that of today's computers, and much of today's work revolves around actually proving that quantum computers can outpace conventional computers in practice.

Golden age of quantum machine learning

Machine learning is another technology -- one not that far removed from the lab -- that may help vault the use of quantum computing forward. Such quantum machine learning is of interest to Google, which is pitted against IBM and others in a breakneck quest to build working quantum computing hardware that may surpass conventional systems.

Google sees search applications, among others, as apt for quantum computing, but it also sees quantum machine learning as a possible pathway to solving computing problems that, for now, seem somewhat unsolvable.

Picture of IBM system for quantum computing
An IBM cryostat chassis wired for a 50 qubit quantum computing system

IBM has a similar impression, and it too is pushing quantum computing research into the area of machine learning. Not coincidentally, Big Blue chose this month's Conference on Neural Information Processing Systems in Long Beach, Calif. -- an event focused on machine learning and computational neuroscience -- to display an early model of its gold-plated superconducting qubit system. No, the booth swag didn't include qubits of zircon, but users might eventually find analytics gold with the system.

For 1QBit's Samli, the application of quantum computing to machine learning makes sense. He pointed out that much of the work of machine learning today is preprocessing -- large-scale preprocessing.

Much of the work of machine learning today is preprocessing -- large-scale preprocessing.

That means working to narrow the deepest processing to the most relevant data. It's a basic principle familiar to any data manager. But it becomes a bigger and bigger problem as the data sets get bigger and bigger, which is part and parcel of big data machine learning these days.

"Before we can effectively do machine learning, we have to first analyze big data. The machine learning sets tend to be highly dimensional data sets. You have to reduce the complexity of the raw data for further computing," Samli said.

Still, Samli cautioned that recent quantum advances shouldn't obscure the fact that there is much more work ahead before quantum computing has broad commercial use.

According to Samli, 1QBit's approach is to build software that isn't tied to any single type of quantum hardware and to support quantum simulations on conventional hardware, as well. For quite some time to come, and more and more going forward, he said, a hybrid combination of the conventional and the quantum will be practiced.

Caution on quantum machine learning is in order. The obstacles to overcome for quantum computing in general include error correction, the persistence of the qubit -- referred to in quantum circles as its coherence time -- and verification of quantum algorithms, not to mention the many vagaries that attend the operations of the low-temperature superconducting systems that represent most quantum computing activity to date.

But, with quantum computing, it may be time to temper caution with some curiosity, especially given the role that quantum technologies could play in expanding analytics on large data troves. 

Dig Deeper on Machine learning platforms