I'm happy that the title doesn't call the D-Wave a quantum computer (as is frequently done), as it's not a quantum computer in the sense that people mean when they say quantum computer. Unfortunately D-Wave is referred to as a quantum computer within the article several times. D-Wave have called themselves a quantum computer for such a long time (prompting many others to call them out, e.g. Scott Aaronson [1], VentureBeat [2], Physics World [3], Ars Technica [4], Wired [5], and on and on) that I am now extremely skeptical of any result that features D-Wave.
The last time I checked (which was a few years ago), it was much faster to simulate magnetic annealing on classical machines than to use the D-Wave --- and this is precisely what D-Wave is designed to do. I wonder whether this has changed? It is hard to quickly check this now amidst the large amount of articles decrying D-Wave for calling themselves a quantum computer.
Towards the end of this article it is noted that several "classical" machine algorithms were also used and these produced essentially the same results.
I think the question of whether one could somehow perform better or faster machine learning, or to write a 'quantum evolutionary algorithm' that perhaps converges to something better faster, is an interesting question. I haven't thought about this before.
Indeed, and I think it's quite telling that the founder of D-Wave, along with some former D-Wave employees, have given up on their vision of quantum computing and are focusing their efforts on making industrial robots instead.
It's true that there is a distinction between a quantum computer build using a universal set of gates(what you probably mean), and one that uses the adiabatic model to carry out computations. The former, is difficult to construct due to decoherence, while the latter carries out computations in the ground state, neatly sidestepping the issue. Saying the D-wave computer is not a quantum computer simply because it does not use a universal set of gates is patently false, it uses a different, more restricted model true, but one that still exploits the resources of the quantum realm to carry out computations. It's also true that theoretically, the adiabatic model is less developed, and thus supremacy arguments are more difficult to make, but as long as quantum effects(i.e. superposition, entanglement, etc) are used to carry out computations, it is a quantum computer.
> The last time I checked (which was a few years ago)
The D-Wave machine is not a full adiabatic quantum computer, though (which would be polynomially equivalent to a universal quantum computer). It does quantum annealing over a pretty restricted set of Hamiltonians.
I think it is perfectly reasonable to say it is not a quantum computer, in the same way that not everything performing computation is called a computer. Otherwise---taken to the absurd---why not call a chemist's beaker a quantum computer?
It's a computing machine taking advantage of certain quantum effects to do computing. Theoretically, it may be able to do certain tasks faster than a traditional computer
I'm not saying you're wrong- it's often mislabeled as a quantum computer- but it isn't fair to say it's no better than a chemist's beaker.
"The D-Wave system uses quantum annealing, which utilizes quantum effects to solve a select few problems much faster than traditional computers. However, it's not as flexible as a universal quantum computer, which could solve a much wider range of problems, including the prime-factorization problem at the heart of modern cryptography."
That's a pretty good summary, but honestly the vast majority of journalists will probably just run with the more sensational headline, and D-Wave is very happy to encourage that.
Is instruction pipelining a form of multiprocessing? (No.) Is hyperthreading? (Sort of, but it's limited and has a separate name for a reason.) Are hyperthreads marketed as SMP cores? (Of course.)
Is a chemist's beaker a quantum computer? (No.) Is D-Wave's device? (Sort of, but it's limited and should have a separate name.) Is D-Wave's device marketed as a quantum computer? (Of course.)
The whole point of a "quantum computer" is getting those nonlinear speedups. If the entanglement doesn't speed anything up then it's pretty close to the beaker in terms of relevance.
Remember, what we are talking about is expressly using the features of quantum theory, superposition, entanglement, etc as a computing resource. This is not at all the same thing as when people say things like 'we design computers with quantum theory' or the like.
A) Uses quantum effects like superposition and entanglement to compute, no quantum speedup.
B) Uses quantum effects to compute, with quantum speedup.
C) Universal quantum computer with quantum speedup.
I would argue that the term "quantum computer", as commonly used, requires the device to be at least class B on this list. Even if you don't think class C is necessary, class A is full of trash like "knockoff 6502 with a fancy photon-splitting ALU". Class A allows the quantum nature to be a pointless novelty resource. It's like a "solar-powered" house with one square foot of panels.
And as far as I know D-Wave haven't conclusively proven their device to be class B.
I don't believe they've proven class A. Just because you have a device that computes something and that devices has entanglement does not mean you've exploits the entanglement to compute anything in any useful sense. An extreme example: you can prepare a superposition and measure it to generate a random classical bit (0 or 1), and you can use that as a source of entropy for a classical probabilistic algorithm, but it would be ridiculous to call this a "quantum computer" of any class.
In fact, multiple experimental groups did essentially this, which some theorists at IBM (former colleagues of mine) called them out for:
> Normal transistors are using quantum effects too.
Well, I mean, I think it's more like "we're well aware of quantum effects and they're parasitic". I've yet to see a production TFET or something like that which actually uses a quantum phenomenon as the operating principle of the switch.
but you don't need to understand much to just physically exist. Whereas electrodynamics and certainly also quantum effects in semiconducting cristals informed the development of transistors in the last what, 50-60 years.
a tunnel diode uses quantum effects, too, and the tiniest mosfets come close to the quantum realm at least in production, too. Does that make a pc a quantum computer? I really don't know the difference, but I'm sure that would be stretching it.
If it's not fighting decoherence then it's not what we mean when we talk about quantum computing. Otherwise I can start saying "technically, you're using a quantum computer right now".
>> It is hard to quickly check this now amidst the large amount of articles decrying D-Wave for calling themselves a quantum computer.
Tip for quickly getting past all the irrelevant results in Google when you're searching for something specific: click the 10th page of results and start from there.
You'll trigger their bot check more often because it's unusual to them that you actually want to know about the specific thing you're looking up instead of seeing 100 identical blog posts/news articles on the subject.
Sorry, but could you go into a bit more detail on how triggering a bot check is supposed to help get past the irrelevant results? I'm not sure I understand.
28 comments
[ 4.1 ms ] story [ 29.1 ms ] threadThe last time I checked (which was a few years ago), it was much faster to simulate magnetic annealing on classical machines than to use the D-Wave --- and this is precisely what D-Wave is designed to do. I wonder whether this has changed? It is hard to quickly check this now amidst the large amount of articles decrying D-Wave for calling themselves a quantum computer.
Towards the end of this article it is noted that several "classical" machine algorithms were also used and these produced essentially the same results.
I think the question of whether one could somehow perform better or faster machine learning, or to write a 'quantum evolutionary algorithm' that perhaps converges to something better faster, is an interesting question. I haven't thought about this before.
[1]: https://www.scottaaronson.com/blog/?p=1400 [2]: https://venturebeat.com/2013/12/27/d-wave-a-multimillion-dol... [3]: http://physicsworld.com/cws/article/news/2014/jun/20/is-d-wa... [4]: https://arstechnica.com/science/2016/02/is-d-waves-quantum-p... [5]: https://www.wired.com/2014/05/quantum-computing/ [6]: https://www.nature.com/nature/journal/v550/n7676/full/nature...
D-Wave has/had several co-founders. One has left to start Kindred.
> The last time I checked (which was a few years ago)
A lot has changed.
Could you please give some examples?
I think it is perfectly reasonable to say it is not a quantum computer, in the same way that not everything performing computation is called a computer. Otherwise---taken to the absurd---why not call a chemist's beaker a quantum computer?
I'm not saying you're wrong- it's often mislabeled as a quantum computer- but it isn't fair to say it's no better than a chemist's beaker.
"The D-Wave system uses quantum annealing, which utilizes quantum effects to solve a select few problems much faster than traditional computers. However, it's not as flexible as a universal quantum computer, which could solve a much wider range of problems, including the prime-factorization problem at the heart of modern cryptography."
Is instruction pipelining a form of multiprocessing? (No.) Is hyperthreading? (Sort of, but it's limited and has a separate name for a reason.) Are hyperthreads marketed as SMP cores? (Of course.)
Is a chemist's beaker a quantum computer? (No.) Is D-Wave's device? (Sort of, but it's limited and should have a separate name.) Is D-Wave's device marketed as a quantum computer? (Of course.)
The whole point of a "quantum computer" is getting those nonlinear speedups. If the entanglement doesn't speed anything up then it's pretty close to the beaker in terms of relevance.
A) Uses quantum effects like superposition and entanglement to compute, no quantum speedup.
B) Uses quantum effects to compute, with quantum speedup.
C) Universal quantum computer with quantum speedup.
I would argue that the term "quantum computer", as commonly used, requires the device to be at least class B on this list. Even if you don't think class C is necessary, class A is full of trash like "knockoff 6502 with a fancy photon-splitting ALU". Class A allows the quantum nature to be a pointless novelty resource. It's like a "solar-powered" house with one square foot of panels.
And as far as I know D-Wave haven't conclusively proven their device to be class B.
In fact, multiple experimental groups did essentially this, which some theorists at IBM (former colleagues of mine) called them out for:
https://www.nature.com/nature/journal/v499/n7457/full/nature...
Although this case was egregious, more subtle shenanigans are rampant in the field.
Well, I mean, I think it's more like "we're well aware of quantum effects and they're parasitic". I've yet to see a production TFET or something like that which actually uses a quantum phenomenon as the operating principle of the switch.
In the same way you're being supported by electromagnetic forces when you're standing or sitting on something
He was referring to D-Wave not being competitive with conventional systems.
When did it become generally accepted that D-Wave has shown it could do anything useful significantly faster than a regular computer?
Last I heard it was only coming close on 1 or 2 things, but still with no convincing advantage let alone a dramatic one,
Tip for quickly getting past all the irrelevant results in Google when you're searching for something specific: click the 10th page of results and start from there.
You'll trigger their bot check more often because it's unusual to them that you actually want to know about the specific thing you're looking up instead of seeing 100 identical blog posts/news articles on the subject.