Quantum Computer Learns To 'See' Trees (sciencemag.org) 116
sciencehabit writes from a report via Science Magazine: Scientists have trained a quantum computer to recognize trees. That may not seem like a big deal, but the result means that researchers are a step closer to using such computers for complicated machine learning problems like pattern recognition and computer vision. The team fed hundreds of NASA satellite images of California into a D-Wave 2X processor, and asked the computer to consider dozens of features -- hue, saturation, even light reflectance -- to determine whether clumps of pixels were trees as opposed to roads, buildings, or rivers. They then told the computer whether its classifications were right or wrong so that the computer could learn from its mistakes, tweaking the formula it uses to determine whether something is a tree. After it was trained, the D-Wave was 90% accurate in recognizing trees in aerial photographs of Mill Valley, California. The results demonstrate how scientists can program quantum computers to 'look' at and analyze images, and opens up the possibility of using them to solve other complex problems that require heavy data crunching.
wow (Score:5, Funny)
Re:wow (Score:5, Funny)
Number Three. The Larch.
The Larch.
Re: (Score:1)
And now for something completely different...
Re: wow (Score:1)
It's not treemendously useful without letting us know how the average human compares on a set of controlled tests.
Re: (Score:2)
It's not treemendously useful without letting us know how the average human compares on a set of controlled tests.
I think most humans can say whether something is a tree or a box of rusty car parts fairly easily.
Careful discrimination is called for (Score:2)
Well, you might think so. But it's not always as easy as all that. [hotrod.com]
Re:wow (Score:5, Funny)
I agree. But can it recognize a forest through those trees?
Re: (Score:1)
I agree. But can it recognize a forest through those trees?
I came here for this joke. Was not disappointed. :)
Re: (Score:2)
Here's another [xkcd.com].
When the user takes a photo, the app should check whether they're in a national park...and check whether the photo is of a bird.
Hmmm (Score:1)
Does dead from the neck up count?
Because there are a lot of pictures of Donald Trump that could be used to train an image recognition system to do that.
Re: (Score:2)
If God didn't intend for men to be fried in oil why did he attach eggs and sausage to them ?
Points... (Score:1)
https://www.bing.com/search?q=... [bing.com]
Re: Yes, but... (Score:1)
If it's a 5-bit quantum computer, only if those trees are five pixels or less.
I bet it still can't (Score:3)
Re: (Score:1)
Re: (Score:1)
I saw the headline and immediately came to the comments looking for this post. Thank you for not disappointing :)
Re: (Score:2)
I would think it could use branching logic to figure it out.
Annealing again (Score:5, Informative)
The 2X DWave machine does quantum annealing. This is good in the sense that it doesn't come remotely close to the kind of machine that maintains coherence across hundreds of bits that can mess up public key crypto. It managed 5 qbits for 20us if read the paper right.
The paper proudly points out it managed 9% error compared to 10.5% error rate of a classical computer doing simulated annealing. However this is not better than classical computer running a better distinguisher. Classical computers are not constrained to run only simulated annealing.
We have yet to reach the point where any quantum computer is faster than a classical computer at any task. This is a good thing. No one has really put anything in place to protect the cryptography of commerce from a hypothetical but largely impossible quantum computer running Shor's algorithm or Grover's algorithm. New hope, RWLE, hash based signatures etc are still the domain of IACR papers. You bank will not be using them any time soon. Lattice crypto just keep getting broken.
Re: (Score:3)
This is good in the sense that it doesn't come remotely close to the kind of machine that maintains coherence across hundreds of bits that can mess up public key crypto.
I don't know, I consider it a bad thing. It would cause problems, but there would be huge benefits too, and the problems could be solved.
Re: (Score:2)
This is good in the sense that it doesn't come remotely close to the kind of machine that maintains coherence across hundreds of bits that can mess up public key crypto.
I don't know, I consider it a bad thing. It would cause problems, but there would be huge benefits too, and the problems could be solved.
Which problems could be solved that would yield huge benefits when solved?
Re: (Score:2)
Re: (Score:3)
That article supports my assertion that there aren't many compelling quantum algorithms unless you are trying to break crypto.
Lattice crypto mentioned in the article is not a quantum algorithm. It's a class of classical algorithms, some of which claim to be secure against quantum computers. They are also demonstrably hard to get right so they are secure against normal computers.
The paper it references "Exponential Lower Bounds for Polytopes in Combinatorial Optimization" is about optimization problems, whic
Re: (Score:1)
Re: (Score:2)
The comment is new. The state of affairs isn't.
Re: (Score:2)
Simulated annealing is a tradeoff of speed and accuracy. It can be proven to find the optimal solution if you're willing to take long enough, but that's usually way too long. (SA is not a way around NP-hardness, so a guaranteed optimal solution takes at least as much time as any more conventional method.)
Re: (Score:2)
I cannot agree with you.
First, it's not unreasonable to assume that intelligence agencies with high budgets already have quantum computers that can break current public key crypto, especially RSA. (There are even rumors that 1024 bit RSA is broken conventionally in actual practice.) After all, we're talking about agencies who are supposed to work on this since the 70s and are among the largest employers of mathematicians and engineers world-wide.
Second, advances in civilian quantum computing are a good thin
Re: (Score:2)
(1 & 2) Yes. This is a nit I have with people who describe quantum interactions as reversible, because you can run the equations for the wave equations backwards. You can, as long as you ignore every other bit of energy in the universe but everything we know tells us you can't ignore thermodynamics when it comes to causality. Things are not reversible. That we have failed to implement useful quantum computers one data point consistent with this view.
(3) I hadn't considered side channels on quantum comp
Re: (Score:1)
This isn't a matter of enough money. It's a matter of theoretical possibility and astronomical technical [hurdles]...
To be fair, it's also a matter of enough money and other resources. That is, even if we assume the NSA has magically solved all the technical problems (which I, for one, think is hugely unlikely), and has a QC that can break, say, 2048-bit RSA using Shor's algorithm. How long does it take to do that? How many private keys can it derive in practice, in unit time? Shor's algorithm isn't instantaneous - it's fast, in complexity terms (log N), but it does take some time. And setup and extraction are also going
Re: (Score:2)
I'm not sure in what way we disagree. My comments were all to do with the fashion in which people's response to the cryptographic threat presented by quantum computers is inadequate and will remain to be inadequate - all assuming such things are possible. Banks certainly should adopt quantum safe algorithms, but they haven't even got past 3DES yet.
I probably disagree that there are benefits to quantum computers if they exist, at least of a type that counterweights the dis-benefits of breaking a lot of in-th
Re: (Score:1)
Re: (Score:2)
Yes. I wasn't saying it's not useful or interesting. I'm saying the benefits of speeding it up a bit are tiny compared to the costs of a quantum computer.
Re: (Score:2)
>this article here is a bit of a scientific popularization but gives a hint on what I'm talking about : >http://www.cs.virginia.edu/~robins/The_Limits_of_Quantum_Computers.pdf
Yes. I read his book. He doesn't dumb it down.
Re: (Score:1)
You jest, but every time I'm asked to give a speech on a complex subject the consistent feedback that I get is that I need to dumb it down even further.
I think I'm going to start a new public speaking career where I deliberately make people feel stupid to punish them for not educating themselves. Asking public speakers to pander to illiterates is the same sort of body positivity nonsense which is encouraging fat chicks to wear skin tight clothing.
Re: (Score:2)
He speaks the truth. A person's desire to understand is often bounded by their willingness to accept a deep thought into their head.
Re: (Score:1)
Round and round we go... hey! (Score:2)
And by "we", you mean, "some people", because there are (and probably always will be) a whole bunch of us out here who just laugh when some committee tries to claim that Pluto isn't a planet.
Planet is an arbitrary descriptor. Pre- or post-committee. I'll keep my descriptor; they can keep theirs.
FWIW (which is something to me, and likely nothing to anyone else) this is the heart of my descriptor:
If i
Re: (Score:2)
Because the most useful concepts describe what you're looking at in such a way as you get an accurate perception of what they are. The current non-planet trend for Pluto doesn't do that for me.
If you follow the latest "trends", Pluto is trending back [usra.edu]...
Unfortunately, as part of this trend, "that's no moon orbiting around the earth, that's a planet..." And we get 110 "planets"...
All in the spirit that everyone gets a participation medal ;^)
Re: (Score:2)
I go with where the centroid of gravity is. If we have A and B that are both large enough to pull themselves into spheroids as above, and they are orbiting each other, but the centroid is inside B, then A is a moon and B is a planet. If the centroid is inside A, then B is a moon, and A is a planet. If the centroid is outside both, then both are planets and we have a sister planet system.
Re: (Score:2)
I've decided that, for my personal definitions, Pluto is a planet, a planet being something big enough that gravity makes it spherical and not so big as to do its own internal fusion. (Fusion reactors on the surface don't count.) I figure there's eight major planets in the Solar System.
Re: (Score:1)
Color me unimpressed (Score:1)
They trained it to identify green pixels? How groundbreaking.
Re: (Score:3)
Dude, it's a quantum computer. It can identify every shade of green!
Quantum look like (Score:2)
Positronic brain in Asimov's novel!
Re: (Score:1)
Finally we will know... (Score:2, Interesting)
..the answer to the most important question in quantum uncertainty: "if a tree falls in the woods and nobody is around to hear it, does it make a sound?"
Re: (Score:3)
well, it does ..... and it doesn't
Re: (Score:2)
That'e because there was... and wasn't... a cat in that tree.
Tree Felling Robots not there yet (Score:3)
90% is NOT good (Score:2)
90% sounds good, but that's also a 10% failure rate. Ewwwwwwwwww, not good.
It's a good start but I wouldn't trust anything that's wrong 10% of the time. When it gets to 99.999999999%, then I'd trust it.
I agree, detecting trees may not be a life or death problem, but then again it might if you're landing a plane or something similar. But it's a good start and will probably only get better.
Re: (Score:1)
90% sounds good, but that's also a 10% failure rate. Ewwwwwwwwww, not good.
That's not how it works. The goal in this kind of situation is not to get a perfect score, it's to get a somewhat reliable one that can be used as an indicator in a larger decision process.
It's like being the owner of a convenience store and seeing a potential customer walking around the aisles. If;
1) the person smells like piss and sweat
2) the person wears multiple layers of mismatched Salvation Army clothes
3) the person is engaged in a conversation with an invisible counterparty
4) the person has shit stai
Re: (Score:2)
90% sounds good, but that's also a 10% failure rate. Ewwwwwwwwww, not good.
That's not how it works. The goal in this kind of situation is not to get a perfect score, it's to get a somewhat reliable one that can be used as an indicator in a larger decision process.
It's like being the owner of a convenience store and seeing a potential customer walking around the aisles. If; 1) the person smells like piss and sweat 2) the person wears multiple layers of mismatched Salvation Army clothes 3) the person is engaged in a conversation with an invisible counterparty 4) the person has shit stains on their shoes
then even if every single one of these observations is only 90% reliable, the big picture clearly tells you that you're dealing with a homeless person and should pay attention to what he's doing.
By a simple application of the laws of statistics, with four separate data points, you can be 360% certain.
Re: (Score:2)
90% sounds good, but that's also a 10% failure rate. Ewwwwwwwwww, not good.
That's not how it works. The goal in this kind of situation is not to get a perfect score, it's to get a somewhat reliable one
Tell that to the pilot landing your plane at night in the fog. I do NOT want him to be "90% sure" that those aren't trees in front of us.
Re: (Score:2)
It's better than Marc Bolan.
What? Too soon?
Re: (Score:2)
It's better than Marc Bolan.
Flashback triggered...
Hard to believe he's been gone for 40 years. :(
Jesus christ, I *am* old.
Re: (Score:2)
Not that bad if you read the article and noted it was high altitude aerial views and in terms of recognition of trees everyone would fail. There is not detail on resolution of images taken or multiple images stitched together or altitude and based on the images no attempt was made to recognise divergent conditions, autumn, substantial different trees types and even shrubbery versus trees. The were not ever clear on with it differentiated between trees or clumps of trees.
Fractal pattern recognition seems th
Re: (Score:2)
How well does it compare to humans?
We'd probably not get a 100% score either.
Re: (Score:2)
How well does it compare to humans?
We'd probably not get a 100% score either.
I bet the average human would do a hell of a lot better than 90%.
A revolutionary development... (Score:5, Funny)
Re: (Score:2)
This will force CS programs to reevaluate how they teach about tree structures. Why traverse a tree when you can look at it?
Because in quantum computing, as soon as you observe the tree it becomes a single fixed state and the whole thing just sits there doing nothing until you look away.... Kinda like the Weeping Angels in Dr Who...
Re: (Score:1)
Why did a CS student traverse a tree?
Re: (Score:1)
No, because the chicken wouldn't do it.
How is it better? (Score:2)
While interesting, the obvious thing they seem to be leaving out is the performance comparison to a regular processor of similar complexity. I'm not trying to pit a 1152 qubit machines against one with trillions of transistors, it would just be nice to have some semblance of a comparison for the reader.
CYA (Score:1)
hue
It's not racial profiling when the quantum computer says YOU'RE A TERRORIST.
Call me when it sees dead people. (Score:2)
Or dead pixels [slashdot.org]. There's work to be done.
I've seen something like this before (Score:2)
A friend was doing image processing work in the late 80s and managed to get some funding for image recognition test. The idea was an alarm camera could detect the family pet and ignore it but detect bad people and set off the alarm. The system was trained using photos of dogs. The end result was a program that could identify pictures of dogs. It was hopeless at detecting real dogs, but it was spot on about detecting pictures of dogs.
Re: (Score:2)
So in order to fix the problem, they had the systems' digital camera take a photo every 30th of a second and it was spot on again?
Quantum Weakness (Score:3)
This story would be a whole lot more meaningful if, for example, the resulting 90% performance of these quantum chips is any way could be contrasted with ANY frame of reference to reality. Whats the point of reporting test results as compared to ...what? What the hell does it mean?... that every not quantum computer can't perform at all? If I am supposed to care about these results compared to nothing at all then I'm in grave doubt as to the value of this technology in the hands of those who can't distinguish it from an iPhone app's performance. I'm pretty sure it can recognize a tree too. Quantum weakness in rhetorical value. Wake up the other hemisphere on this half baked argument so that nerd news can matter to the rest of the human race.
I imagine this is the training program (Score:5, Funny)
how much did it really learn (Score:1)
If the picture was taken in fall or winter when you have multi colored or no leaves, does it still work?
About the trees that it can observe (Score:1)
Quantum? (Score:2)
I thought in quantum computing a tree could be a tree or not, or both a tree AND not a tree. So how can we be sure what it's telling us?