Slashdot Mirror
Baidu's Supercomputer Beats Google At Image Recognition
catchblue22 writes: Using the ImageNet object classification benchmark, Baidu’s Minwa supercomputer scanned more than 1 million images and taught itself to sort them into about 1,000 categories and achieved an image identification error rate of just 4.58 percent, beating humans, Microsoft and Google. Google's system scored a 95.2% and Microsoft's, a 95.06%, Baidu said. “Our company is now leading the race in computer intelligence,” said Ren Wu, a Baidu scientist working on the project. “I think this is the fastest supercomputer dedicated to deep learning,” he said. “We have great power in our hands—much greater than our competitors.”
I'm not sure an improvement of .5 percent on image cataloging is really that impressive to get not one but two greats...
This is actually News for Nerds.
I'm curious how much difference in computational power was thrown at training these by Google, Microsoft, and Baidu, though it's going to be great to watch how these continue to evolve.
Only $7.42 including shipping! (AC adapter not included.) Estimated 21-47 days delivery time to USA.
After all the news stories from the past couple of years, it seems like you could just guess "Yeah, that's a penis" and be correct about half the time. Seems like most people if you give them a camera, they're going to take a picture of a penis with it. And subsequently post that picture to the internet somewhere.
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
The summary is written to imply that Google/MS have error rates in the 90's, while the competition only has about 5% error. The values got inverted - Google/MS also have error rates around 5%, but are behind by fractions.
As a pedant, I need to point out that the improvement is 0.24%
Also why are the numbers reversed to quote success rates for Google and Microsoft in the summary on Slashdot - it would have been much clearer if the actual numbers in the article (which were all error rates) were quoted!
congrats on the sensationalism.
Okay, so we have a benchmark where the bog-standard human being scores 94.9%.
Then in February (that's three months ago), Microsoft reports hitting 95.06%; the first score to edge the humans.
Then in March, Google notches 95.18%.
Now it's May, and Baidu puts up a 95.42%.
Meh. Swinging dicks with big iron are twiddling with their algorithms to squeeze out incremental, marginal improvements on an arbitrary task.
“Our company is now leading the race in computer intelligence,” said Ren Wu, a Baidu scientist working on the project. ... “We have great power in our hands—much greater than our competitors.”
I presume that next month it will be IBM boasting about "leading the race" and being "much greater than their competitors". The month after that it will be Microsoft's turn again. Google will be back on top in August or so...unless, of course, some other benchmark starts getting some press.
~Idarubicin
Build a "Watson style" chatterbot that can win on Jeopardy and despite this miraculous achievement, have the company go under because management are dicks to the level of being able to fuck up a free lunch with an error rate of just 4.3%.
Not quite...
The real question, of course, is whether Google, Microsoft, and Apple will soon have to face a serious international competitor. It's true that Baidu's incremental image recognition changes might not be a game changer. But if there's any substance to these claims about speech recognition, Baidu might be on track to produce an actual competitive advantage in ways highly relevant to consumers.
beating humans, Microsoft and Google.
Sure, but not beating several gov't intel organizations around the world... likely.
But can they claim that they are better than the NSA at image recognition? Somehow I doubt it.
Baidu said "Your Kung Fu no good in my village"
"If any question why we died, Tell them because our fathers lied."
I only took basic AI in university but...
The power of the computers is not the important thing here if it takes 3 weeks to train the neural network or 1 day does not change the ACCURACY. The running of the NN to identify a picture is also only a fraction of the training time.
Maybe its more about the TRAINING SET here rather than CPU power. It seems extraordinary. 1 000 000 images sorted into 1000 categories must have been done by humans right? Humans sitting like dog,dog,dog,airplane,dog, house, dog OHPLEASELETMEJUSTDIE.
A bigger trainingset is usually a good thing with NN to avoid overtraining which makes it generalize worse on test examples.
The computer has 72 processors and 144 GPU's. That's tiny. Seriously tiny. Sure, GPU's are powerful, especially for image processing. But the larger computers these days are running tens to hundreds of thousands of processors in parallel.
For example, assuming each shelf has 2 processors and 4 GPU's, and they can fit 12 shelves into a single rack, that's a total of 2 racks. Compare that to this image of one of Google's datacenters, where you can see dozens of racks, each containing 14 shelves by my count. And that's just one row. These are gigantic warehouses, with row upon row of racks.
The level of processing power claimed here is closer to the level of a university processing cluster. The larger scientific clusters can be ten or a hundred times larger, and it's not clear just how big private datacenters are.
So overall I'm very, very skeptical. There's a very good chance that they fudged the data somehow to make theirs appear better. But if it is better, well, there's no reason why Google and Microsoft couldn't easily outcompete them in short order.
As you near 100% success, the ratio of failures is much more important than the difference in percentage.
For example, think of the case where two things have 99.75% and 99.99% success (i.e. 0.25% and 0.01% failure). It's the same difference of 0.24%, but it's much more significant because they're closer to 0% failures. In this fictional example, the one with 0.25% failure has 25x as many failures as the one with 0.01% failure!
tl;dr: 4.58% / 4.82% = 0.950, or 5% fewer failures.
I wonder how good farce book is at this?
I was fantasic http://denkepbanpixar.blogspot...
It may not be processing power alone, but perhaps a better learning algorithm.
When it comes to solving problems, elegance can sometimes beat out brute force. :D
Marginal improvements are worthwhile if the cost of failure is relatively large. c.f. branch prediction in computer architecture....
Baidu doesn't have as many women or as much racial diversity, which is why they're going to fall behind pretty quickly.
There's nothing really wrong with this announcement -- It's just not a big breakthrough of any real sort.
OS Software is like love: The best way to make it grow is to give it away.
The computer has 72 processors and 144 GPU's. That's tiny. Seriously tiny. Sure, GPU's are powerful, especially for image processing. But the larger computers these days are running tens to hundreds of thousands of processors in parallel.
For example, assuming each shelf has 2 processors and 4 GPU's, and they can fit 12 shelves into a single rack, that's a total of 2 racks. Compare that to this image of one of Google's datacenters, where you can see dozens of racks, each containing 14 shelves by my count. And that's just one row. These are gigantic warehouses, with row upon row of racks.
The level of processing power claimed here is closer to the level of a university processing cluster. The larger scientific clusters can be ten or a hundred times larger, and it's not clear just how big private datacenters are.
So overall I'm very, very skeptical. There's a very good chance that they fudged the data somehow to make theirs appear better. But if it is better, well, there's no reason why Google and Microsoft couldn't easily outcompete them in short order.
This is not a competition of hardware. It is a competition of software and algorithms. The hardware is somewhat irrelevent, and a smaller hardware footprint would actually be more impressive! You also seem to think that the #1 search provider in the chinese market is incapable of standing toe to toe with an American company. I think you need to get out a little more.
Nix...
Obligatory link to Richard Pryor's famous joke about big dicks.
If the #t=470 doesn't work, just skip to 7:50. The joke only takes 50 seconds. You won't be disappointed, even though you've probably heard it before. ;)
“We have great power in our hands—much greater than our competitors.”
Pffff. I hope it's just a translation artifact, because it sure sounds ridiculously childish.
Nah; next will be Wolfram, based on crowdsourcing.
You said (emphasis mine):
But the larger computers these days are running tens to hundreds of thousands of processors in parallel.
I don't know what GPU they're using, but if they're Nvidia GeForce GTX Titan Z (700 series; released in March 2014), then that could be 144 * 2880 * 2 = 829,440 shader cores, obviously in parallel (running at 705MHz to 876MHz).
That card can do 8.1 TFLOPs (single-precision), or 2.7 TFLOPs (double-precision). That means 144 of them could do over over 1.1 PFLOPs (single-precision). That's nothing to sneeze at.
p.s. I got my figures from here.
More info on the specs of the "supercomputer" that TFA only glossed over:
The result is the custom-built supercomputer, which we call
Minwa . It is comprised of 36 server nodes, each with 2
six-core Intel Xeon E5-2620 processors. Each sever con-
tains 4 Nvidia Tesla K40m GPUs and one FDR InfiniBand
(56Gb/s) which is a high-performance low-latency inter-
connection and supports RDMA. The peak single precision
floating point performance of each GPU is 4.29TFlops and
each GPU has 12GB of memory. Thanks to the GPUDirect
RDMA, the InfiniBand network interface can access the re-
mote GPU memory without involvement from the CPU. All
the server nodes are connected to the InfiniBand switch.
Figure 1 shows the system architecture. The system runs
Linux with CUDA 6.0 and MPI MVAPICH2, which also
enables GPUDirect RDMA.
In total, Minwa has 6.9TB host memory, 1.7TB device
memory, and about 0.6PFlops theoretical single precision peak performance.
It's not that powerful overall, but seems to be well thought out for what it is doing. I do the see point about fudging data somehow, they do provide a lot of information of what they supposedly did here
I don't know how this is verifiable, it's not like they have released source code or binaries for the software as far as I can tell.
Marginal improvements are worthwhile if the cost of failure is relatively large. c.f. branch prediction in computer architecture....
Your Mom's relatively large.
When she gets on the bus the ALU just gives up.
Her Branch prediction is that if she climbs branches they'll break off.
And I hear she likes JZ and XOR.
The high end of the TOP 500 super computers use tens of thousands of GPUs (at least among those that use GPUs at all); for instance the Titan at ORNL has 18,688 nVidia Tesla K20's for a total of (roughly) 46 million CUDA cores.
One generally does not count individual CUDA cores, however (nor the equivalent for AMD GPUs).
http://image-net.org/challenge...
Has the 2014 competition, including test images and validation images.
Browsing the images, and the 200 or so categories, "artichoke", "strainer", "bowl", "person", "wine bottle"... the challenge is a bit strange: A drawing of a person isn't a "person" category, but a bottle of boyle's cream soda is a "wine bottle".
And why is "artichoke" something we need to identify in photographs?
Sure, but that's why Microsoft and Google will rapidly catch up if the numbers are real. Both employ lots of extremely talented and creative people exactly for solving problems like this, and the methods they use have been published.
Anyway, if they did really manage to produce some better algorithms, that's impressive and important work. But bragging about such a tiny computer seems seriously out of place.
And you get +4 for claiming an algorithm improvement is a 'trick'?
Don't be too proud of this technological terror you've constructed. The ability to recognize an image is insignificant next to the power of the Force.
a) Each CPU in these clusters typically has anywhere from 4-8 cores, and may support two or more times as many threads.
b) It's far, far more difficult to make full use of GPU hardware than CPU hardware. The best application for stressing GPU hardware is 3D graphics rendering, and even there if you run through the numbers, you find that it's rare that they really push half of their theoretical processing limit. General processing is significantly less efficient on GPU hardware, in particular because it's difficult to come up with computing problems that work well with the GPU's extremely limited I/O compared to their processing power. You need to do a lot of processing on each bit of data read or written to not be limited by either PCI Express bandwidth or video RAM bandwidth. Typical best-case real world scenarios for GPGPU programming put GPU's at closer to 10x or so the performance of CPU's, not 1000x as just looking at the number of shader cores vs. CPU cores might suggest. So they're quite powerful, but not overwhelmingly so. Whether or not they're worth it is highly dependent upon the application.
c) You can bet that companies like Microsoft and Google have a significant number of GPU's in use for specialized tasks.
they just can't let anyone else "win".
That just makes the real stuff, their algorithms, even more impressive.
And...he's humble.
But if there's any substance to these claims about speech recognition...
Baidu considers a single month advantage consisting of a fraction of a percent better in image recognition as "leading the race in computer intelligence" and being "much greater than our competitors". Due to that alone any claims they make should have a null hypothesis that they are completely full of it.
Until the employ me they will never be able to catch up
Going from 99.5 to 100.0 percent is extremely impressive, while going from 50.0 to 50.5 is probably just noise.
http://xkcd.com/1444/
Do an image search for "man in a purple hat holding a watermelon" Google's results are the most intelligent followed by Bing with Baidu a long way back in third.
First they beat us at math. Then at strategy games. Now they beat us at one of the few things we still did better, visually distinguishing apples from oranges.
Don't waste your vote! Vote for whoever you want, unless you live in a swing state it won't matter anyways
All of them should have their system train on several seconds of video to identify action: drinking, running, etc. Next, train on sequential actions to identify cause and effect. Next, train on movie,tv,news,youtube,cctv audio->text/dialog/script to identify the relationship between words, actions, cause and effect. Finally, they should build a chatbot that maps input words to actions, actions to cause, cause to effects, etc until they get back to output words. It wouldn't pass a turing test but it would be awesome. For example: Human: If I am in a dark room and I turn on the light, what will happen? If I kiss my wife, what will she do? If I get on one knee and give someone a ring, what is happening? It probably would also help with language translation: map english words to actions in english movies, map similar actions in russian movies to russian words.
You do realize WHO is actually working at Baidu right? A Mr. Andrew Ng.
Actually I would surprised if Baidu didn't have much more women at work than either Google or Microsoft since China has state policies promoting gender equality.
Considering most Google data centers are designed with high redundancy in mind, it is hard to use them as a baseline for anything.
Google take "RAID" (RAIC?) to the next level.
It can also easily be something in the software side of things as well.
It will likely be beaten at some point, then they will swing back, then Microsoft will be like, "guis pls slow down", then they will fake numbers because why not it is Microsoft, they did it before for Xbox One cloud rendering.
It is always a constant race of tweaking.
Regardless, it won't matter because humans are terrible at asking for reliable search parameters most times, which is why you get about 50 other unrelated image terms in your searches because they have to assume you mean every context of the word(s) you queried.
But China doesn't have as many women because of the one child policy and female infanticide.
It still mistook a subway carriage for a toilet.
Aren't Asians supposed to be good at math?
So desperate to exaggerate the size of their ePeen
If they "outperformed humans", then what's the basis of the truth set? Wasn't it humans who figured out how the images SHOULD have been categorized?
the fact that the head guy at Baidu now, came from Google. Basically, he took Google's technology and then was funded by China's gov ( who is behind Baidu's funding on this ).
Hopefully, someday soon, the west will realize that hiring Chinese means simply giving your technology over to the CHinese gov.
I prefer the "u" in honour as it seems to be missing these days.
0.24 is about 5% of 4.82
this signature has been removed due to a DMCA takedown notice
Do you want Skynet? Because that’s how you get Skynet.
Get free satoshi (Bitcoin) and Dogecoins
China has state policies on human rights and freedom of religion, as well.
Okay, so we have a benchmark where the bog-standard human being scores 94.9%.
Yes, and now the algorithms are better. More importantly, the 'standard human' only does that when it is paying attention, which it can't do for more than 15 minutes or so. The computer does it day in, day out, forever. And it will get better over time.
Then in February (that's three months ago), Microsoft reports hitting 95.06%; the first score to edge the humans. Then in March, Google notches 95.18%. Now it's May, and Baidu puts up a 95.42%. Meh. Swinging dicks with big iron are twiddling with their algorithms to squeeze out incremental, marginal improvements on an arbitrary task.
You denigrate their work, but that's the way science works: incrementally almost all the time. In any field, you will see tweaking, slight improvements, variations, and a couple of new ideas. And then one of the researchers will hit on the next big idea. So what? What the hell have you done? You're just being a dick.
“Our company is now leading the race in computer intelligence,” said Ren Wu, a Baidu scientist working on the project. ... “We have great power in our hands—much greater than our competitors.”
I presume that next month it will be IBM boasting about "leading the race" and being "much greater than their competitors". The month after that it will be Microsoft's turn again. Google will be back on top in August or so...unless, of course, some other benchmark starts getting some press.
First, what they are doing is very hard. So, yeah, doing 0.25% better than someone else is a big deal. Let's see you do better.
Second, look at the performance over time. There was the NIST handwriting sets, and then the Stanford data sets, then the 'standard' was the PASCAL Visual Object Challenge and people were slowly improving to the point that someone else needed to step up and provide a better standard (more categories and more examples of each). And that was the ILSVC, and now we're down to the last couple percent on those. The next set will be bigger and harder. And performance will improve on that one too. That's expected and a good thing. Image recognition is stunningly hard; thanks to the hard work by these researchers it's gotten a lot better.
here's your obligatory XKCD
The more people I meet, the better I like my dog.
Because this is Slashdot and it is required that all stories be written as poorly as possible.
Exactly what I though too:
Meh, they are all basically in the same ballpark (including humans). No breakthrough achievement. Wake me when some one achieve 10x better than humans and competition.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]