The Quantum Sensei: Dr. Hidetoshi Nishimori Discusses Quantum Annealing

Published by whurley (11/13/2017)

Dr. Hidetoshi Nishimori Answers Whurley’s Questions About Quantum Annealing

If you’ve never heard of Hidetoshi Nishimori, then you might as well have never heard of quantum computing. His 1998 theory of quantum annealing has since been studied and tested extensively on real machines, paving the way for much of the progress in the field to date.

Dr. Hidetoshi Nishimori is a Professor at the Tokyo Institute of Technology, and one of the fathers of quantum annealing. He received his PhD from the University of Tokyo, and he did postdoctoral research at Carnegie-Mellon University and Rutgers University. For his work, he has won an IBM Science Prize (1990) and the Nishina Memorial Prize (2006).

I was lucky enough to meet Professor Nishimori at this year’s adiabatic quantum computing conference. It was without a doubt one of the best moments in my quantum career so far. In addition to having a brilliant mind, he is extremely patient and kind when explaining the intricacies of his craft. This week I had the pleasure of asking him some questions about quantum annealing and the current state of quantum computing.

How does quantum computing compare to classical computing? Will quantum computers replace classical computers?

Classical computers will not be replaced in the foreseeable future. Quantum computers will help classical computers solve certain types of very, very hard problems.  Think of a quantum computer like a high-performance race car which runs extremely fast on the track. People don’t usually buy such a car for daily use—shopping, commuting, and whatever else.  Conventional cars, electric or otherwise, will still be on the world’s streets at the end of this century. Classical and quantum computing technologies will play their respective roles in cooperation with each other whenever possible and appropriate.

How would you explain the differences in how quantum gate-model machines and quantum annealing machines solve problems to a layman?

A gate-model quantum computer follows very detailed instructions from the programmer, step by step.  An ingeniously designed algorithm makes it possible to run the machine extremely fast, but the design of such an algorithm requires superior talent. An annealing machine operates under the natural laws of physics without detailed instructions from an external agent.  Imagine a rugged landscape with valleys, mountains, basins, even tunnels between the valleys. After a rainfall, water will naturally run to lower and lower valleys or basins, sometimes through tunnels.  This is similar to what happens with quantum annealing. Once the problem you want to solve is translated into an “energy landscape,” the machine finds the lowest point just by following the laws of quantum mechanics.  The only catch is that you first have to translate your problem to an energy landscape. This is not extremely hard, but requires ingenuity to achieve the best results.

What problems are quantum annealing machines particularly well suited for? Will this change once we have a robust universal quantum computer?

If you want to identify the best possible solution (or something close to it) from among many possibilities, quantum annealing is a good choice.  The taxi-traffic optimization solution Volkswagen created in collaboration with D-Wave is a remarkable example. In  short computation time, they succeeded in finding very efficient routes for hundreds of taxis running from downtown Beijing to the airport on a D-Wave machine. If realized, gate-model (universal) quantum computers with thousands of stable qubits will replace quantum annealing machines. Gate model computers can simulate annealers efficiently. I have no idea when such such a computer will exist, though.

What are the biggest engineering challenges in quantum computing?

It is very hard to keep qubits in stable states in the superconducting technologies currently adopted for quantum computers—annealers as well as gate models—by D-Wave, Google, IBM, Intel, and others. Different quantum computing technologies exist, but they are in nascent stages of development due to the difficulty of controlling the state of qubits precisely as prescribed by quantum algorithms.

Will quantum computing help address the Moore’s law problem?

For some applications, yes. For others, no.  Whichever the type, quantum computers are not a single, unique solution to the coming failure of Moore’s law.  We must develop  other technologies to complement quantum computing’s strengths and weaknesses.

What trends do you see in high-performance computing that will interact with quantum computing?

Quantum computers don’t run alone.  Their capacity—in other words, the number of qubits they hold—will be very limited for quite a while.  We should develop and implement a hybrid system of quantum and classical computing. Data would transfer back and forth between the classical computer and the quantum computer. It doesn’t make sense to use quantum computers for tasks that can be processed efficiently on classical computers.  High-performance computing will take a path toward such a hybrid model, not just with quantum computing, but also with other types of specialized computing models.

How will quantum computing change, influence, or incorporate technologies like artificial intelligence, machine learning, big data, and the cloud?

Most aspects of machine learning are suitable to be processed through quantum annealing. As a result, quantum annealers may well become one of the key tools for training A. I. Provided that a few bottlenecks can be resolved, of course. Most notably the limited number of qubits currently available.  Even with a hundred thousand qubits, it would be impossible to load big data into the system directly to train the A.I.  A hybrid model to operate quantum annealers in tandem with classical computers would be one possible solution, but other ways should also be vigorously pursued.

The gate model quantum computer would be very useful for some machine learning tasks if classical big data can be efficiently transformed to a quantum format. In any case, I am relatively optimistic that quantum computers will be vital to accelerating the development of A.I. to a point that classical computers find hard to reach.

How will the way we currently think about solving problems with a computer have to change?

We will have to choose the right problems to solve. Before you run a quantum computer, you have to consider whether quantum processing is the right tool for your problem. This will be an integral part of “solving” the problem. I think it will be more challenging to find the right problem to be solve than to solve the right problem on a quantum computer.

Are there particular concepts or models from traditional computing that you’ve seen people struggle to leave behind when they’re working with these new quantum machines?

Classical computers always return the same definite output for the same input when processed by the same program.  In contrast, quantum computers run probabilistically.  Outputs can differ from one run to the next. This applies to both the gate model and annealer.  This doesn’t mean that quantum computers are unreliable.  If the probability of getting the correct answer is one out of two, then you run the algorithm twice, and you are very likely to get the correct answer.

What are some of the most prominent myths or misconceptions about quantum computing?

It is a widespread myth that quantum computers will solve any problem at ultrafast speed. Don’t expect that your Nintendo game console will go quantum or your laptop will run Windows Q in your lifetime. Quantum computers show their peak performance only for the right problems, which are limited in number, at least for the moment.

There’s talk that the time is now. What are the challenges that could potentially derail the promise of near term quantum computing?

Scientific research can always go wrong.  We scientists experience more failures than successes. (We don’t tout our failures loudly, though.)  I  wouldn’t promise a rosy future with the advent of quantum computers.  All aspects of the development of quantum computers—hardware, software, and the basic theory—face formidable challenges.  It may turn out to be very difficult to build gate model quantum computers with a capacity beyond a few hundred qubits. Quantum annealing may prove to not be very fast on most practically important problems. Another computing paradigm may suddenly emerge that is much more promising.  I’m more encouraged when the challenges look more formidable; that’s when a breakthrough has a much larger impact.

Was there anyone whose work inspired you to pursue a life in the sciences? Is there someone working in quantum today that inspires you?

Dr. Hideki Yukawa, the first Nobel Laureate from Japan.  The news of his honor in 1949 was a big, big encouragement to all Japanese citizens, young and old, after the devastation of WWII.  I was born in 1954, and my parents and grandmother told me repeatedly that Dr. Yukawa was the pride of Japan. Most Japanese physicists of my age or a little older chose physics as their career under his influence.

What are some of the most interesting projects you’ve seen others working on in quantum computing?

If it can be called a project, Google’s activities in their Quantum Artificial Intelligence Lab are very interesting and inspiring.  They started from scratch several years ago and attracted a lot of young talent, mostly theorists, to extensively test their D-Wave machine. Then they shifted to experiments, hired Professor Martinis strong group out of UCSB, and dove into building the gate-model as well as annealing machines. Today they lead the field on many fronts, from the basic theory to hardware development.  Their approach is flexible and vibrant, involving many theoretical and experimental scientists on all project teams.  Hard for us to match for us, but we can learn a lot from them.

What applications of quantum computing are you most excited for?

Quantum simulation is the single most exciting area to me. It has already been done, to some extent, by a D-Wave researcher, who simulated a quantum magnet of a size that is close to impossible on classical computers.  A next-generation D-Wave machine will show us a world that no other current methods allow us to peek into. As a scientist, I am interested in the working mechanism of magnetism, and very much look forward to seeing the results of quantum magnet simulations.