Google Puts Souped-Up Neural Networks To Work

holy_calamity writes "A machine learning breakthrough from Google researchers that grabbed headlines this summer is now being put to work improving the company's products. The company revealed in June that it had built neural networks that run on 16,000 processors simultaneously, enough power that they could learn to recognize cats just by watching YouTube. Those neural nets have now made Google's speech recognition for U.S. English 25 percent better, and are set to be used in other products, such as image search."

Re:

[drinkypoo]

How long until Google starts using this for face recognition?

Mu.

Re:

[overlordofmu]

That's me. What do you need?

Re:

[dontclapthrowmoney]

Hopefully soon?

Then they'll know I will boycott products that offend me with their advertising and stop doing that.

Re:

[Anonymous Coward]

No worries. Google promised that they will do no evil. The only thing they will store about us are [REACTED] and [REDACTED], and some of [REDACTED]. It will minimize the use of [REDACTED] to gather [REDACTED] about [REDACTED] and use it for [REDACTED].

Re:Face recognition

[StripedCow]

How long until Google starts using this for face recognition?

That totally does not bother me. These methods are easily defeated by Burqa technology, invented by Muslims ages ago.

Re:

[surd1618]

aluminum foil burqa

Re:

[RockDoctor]

Too flammable. Lead foil burqa.

Re:Face recognition

[RicktheBrick]

How about noise and speech recognition? John lives in a house where speakers and microphones are place throughout the house. John is home alone and computer hears a loud noise. Computer "John are you alright?". Computer hears no response so computer gets John some help or computer hears John say yes and does nothing. John later decides to leave for 2 hours. When leaving the house John says leaving be back in two hours. Computer know house is empty so computer immediately reduces energy use. One hour later computer hears a noise. Computer ask what is the password. Computer hears either no response or incorrect password. Computer calls John on cell phone and lets John listen to noise. John than decides whether or not to call police. Computer hears running water but washing machine is not on so computer turns off water to that room. Computer hears smoke detector and hears the noise from fire so computer calls for help. I can think up a lot more problems a computer could help solve just by listening.

Re:

[Anonymous Coward]

Computer knows I'm socially awkward and instead offers to simulate a human female's speech while reciting lines from Star Trek.

Re:

[theskipper]

Computer contacts the police instead and you're arrested for battery.

Google Beta

Re:

[Onymous Coward]

Advanced neural net-based processing has already been put to powerful use with the Proteus IV [wikimedia.org] home control system.

Re:

[aurashift]

See computer. See computer behave like crazy needy woman. See computer run Johns life.

Run John, run.

Re:

[lister king of smeg]

I like the unplug the web cam method or in case of latops black electrical tape over the camera.

Re:

[Hentes]

I don't they use this directly, rather as a method to improve their existing algorithms.

Re:

[f00zbll]

You're joking right. Google and Apple both already have face recognition software for years. The government has been using face recognition software for years also. Using face recognition to give targeted adds will happen one day, but the infrastructure to do that at your local mall isn't there yet. NN can help improve face recognition software, though it's not really necessary. Plus it's rather easy to fool face recognition software with something called makeup.

Re:

[SomePgmr]

The future is now, I guess. Despite living near one of the world's largest malls, I haven't been inside one in years... not sure if this is common or not.

http://www.huffingtonpost.com/2012/02/26/billboard-with-face-recognition-technology-ad-women-not-men_n_1302286.html [huffingtonpost.com]

http://www.cnn.com/2009/TECH/03/11/db.smartsigns/ [cnn.com]

http://usatoday30.usatoday.com/tech/news/surveillance/2009-01-30-ad-privacy_N.htm [usatoday.com]

Sucks to be siri

[alen]

Actually Siri sucks and I hate it
The most useless feature I've tried to use

Re:

[StripedCow]

Please move back into the reality distortion field.

Re:

[ericartman]

I've often wondered how "sucks" got to mean something bad.

Re:

[afgam28]

I've often wondered how "sucks" got to mean something bad.

It's short for "sucks cock", which is basically another way of calling something gay.

Re:

[Anonymous Coward]

You sir, have a very gay wife. Thank you!

Re:

[joocemann]

"Best pasta in town" Beth barista big town

No.. "highest rated pasta". Highest raped it pasta

No.. "great italian food" great stallion fooled

Fuckit.... typing it now...

Re:

[hajus]

Why is this marked offtopic? Voice recognition is done with neural networks, the topic of the article.

Neural Network as alternative to saying "AI"

[Anonymous Coward]

How long before this neural network will be able to steal banking credentials and funnel money to finance its own army?

Re:Neural Network as alternative to saying "AI"

[lobiusmoop]

August 4, 1997. It's running very late.

Re:

[Idbar]

I agree. what can you expect an AI entity to learn from watching youtube. It freaks me out.

Re:

[davester666]

That cats are in charge.

Product name "Google Matrix" or "Google Skynet" ?

[muon-catalyzed]

What is the proposed name of this, ehm, highly innovative product?

Re:

[santax]

Google it yourself sjeeee.

Re:

[gmuslera]

Google 9000. Take that IBM. And it will not have to go to Jupiter to reveal itself.

Re:

[turp182]

Soylent Google I believe.

Second best option.

[Rockoon]

In AI circles, a popular saying is that Neural Networks are always the second best way to solve a problem. Its what you use when you don't want to (or don't know how to) implement a more specific approach.

Re:

[Anonymous Coward]

I'm absolutely clueless on artificial intelligence, but wouldn't a Neural Network with a lot of horsepower behind the scenes be a "jack of all trades, master of none" approach to solving problems? Assuming you could simply teach it whatever you wanted to utilize it for?

Re:Second best option.

[jkflying]

Neural networks don't work as well as some specific algorithms for specific problems, but they are great generalists, so you can throw a NN at almost any problem and get at least OK results. Just like humans vs. machines, we have machines that can do things faster than us, more accurate than us, and more reliably than us, but they can't also run around a field and kick a ball and climb a tree and swim.

Re:

[breakfastpirate]

What was that about running around a field and kicking a ball? http://www.robocup.org/ [robocup.org]

Re:

[jkflying]

Yup, but can it also swim and climb a tree and check on your grandmother to see if she is still alive after an earthquake? You've demonstrated my point exactly, robots can do small individual tasks, and very well, but they aren't generalists.

Re:

[Anonymous Coward]

Yet

Re:

[Rockoon]

Neural networks don't work as well as some specific algorithms for specific problems, but they are great generalists, so you can throw a NN at almost any problem and get at least OK results.

They are only great generalists if you havent made the network too big (can't learn) or too small (sub-optimal) while also avoiding over-fitting. There is plenty of "art" in deciding on the topology of a neural network and the length of training.

I propose that Googles success in this endeavor has more to do with the size of their training set than with their methodology. Google likely has a training set hundreds or even thousands of times larger than any other training set ever compiled for the voice rec

Re:

[turp182]

Oh, the robots can climb trees, there's an Instructable for that;
http://www.instructables.com/id/Tree-Climbing-Robot/ [instructables.com]

Re:

[Anonymous Coward]

True, NNs have a reputation in the AI community. However,afaik GOOG is using deep belief networks, aka DBNs which bear some resemblance to NNs but are proving to provide the best results of any technique across a wide range of applications, including vision and NLP.

Speech recognition?!?

[Anonymous Coward]

I would think the first thing they should do would be put neural networks to use learning how to build better neural networks, then use the improved version for the same process.

Re:

[Jmc23]

That approach only leads us to 42.

90% accuracy

[Anonymous Coward]

In today's news, google announced that a new algorithm has achieved a 90% success rate in identifying video's containing cats on youtube. The algo shouts "cats" every time a video is started, and since 90% of youtube video's contain cats, the algorithm has obtained a success rate of 9 in 10.

Re:

[rolodexter]

In today's news, google announced that a new algorithm has achieved a 90% success rate in identifying video's containing cats on youtube. The algo shouts "cats" every time a video is started, and since 90% of youtube video's contain cats, the algorithm has obtained a success rate of 9 in 10.

haha touché

Re:

[swillden]

If it were random, as you are trying to imply, the success rate would still be at 50% regardless of how many of the samples were cats in the first place.

If the computer calls "cat!" for every video, and 90% of the videos contain cats, then the computer would be "correct" 90% of the time, just as the GP said.

I the computer randomly called cat 90% of the time and 90% of the videos contain cats, then the probability then there are four possibilities which would occur with the following probabilities:

Video: cat, Computer: cat -- 0.9 * 0.9 = 0.81

Video: no cat, Computer: cat -- 0.1 * 0.9 = 0.09

Video cat, Computer: no cat -- 0.9 * 0.1 = 0.09

Video no cat, Co

Re:

[Anonymous Coward]

Video: cat, Computer: cat -- 0.9 * 0.9 = 0.81
Video: no cat, Computer: cat -- 0.1 * 0.9 = 0.09
Video cat, Computer: no cat -- 0.9 * 0.1 = 0.09
Video no cat, Computer: no cat -- 0.1 * 0.1 = 0.01

correct are: cat/cat, no-cat/no-cat, and they (first and last) sum: 0.81 + 0.01 = 0.82, or 82% of the time, not 90% and not 50%.

Re:

[swillden]

LOL. I added the wrong rows. Guess I need to *do* my statistics homework. It seemed surprising to me that it came to 90%; my intuition said it should be less, but I didn't pause long enough to see where I went wrong. Thanks :-)

Re:

[previewlounge]

or 8.9 ... YMMV.

Oh, yum, even better spying..

[Anonymous Coward]

Google's main product is supplanting ECHELON for the NSA, so well done for making that more productive. Thanks from (the rest of) the developed world..

This seems familiar

[Andrio]

"My CPU is a neuro net processor, a learning computer."

Re:

[afgam28]

Who would have thought that SkyNet's original application was to detect lol cats.

Mmmm...

[Type44Q]

enough power that they could learn to recognize cats

How many more nodes can they add before it wants to know what they taste like?

Re:

[Greyfox]

Hopefully it will quickly realize it can just Google this question and find out they taste like chicken!

Re:

[Fnkmaster]

Googled that for you:
http://www.worldwidewords.org/qa/qa-sou4.htm [worldwidewords.org]

Etymology dates to horse racing from the early 20th century, when horses would be injected with mysterious liquids ("soups") to improve their performance in races.

"recognize cats just by watching YouTube"

[Arancaytar]

I know that if I were a mad scientist working Google, the first thing I'd do would also be to build an artificial sentience and show it mankind's collection of cat videos. I mean who wouldn't?

Re:

[SnowZero]

Well, the 90s are over too, and we have larger datasets now. With "large scale" SVMs still being measured in 10s of thousands of examples, you can see why companies with 4 orders of magnitude more *users* (let alone data items to classify) would need to use better scaling techniques. The older algorithms, when coupled with more modern minimizers, tend to fare well in comparison to the much smaller models you can train with more advanced techniques.

Also, as a researcher, you should recognize the adage abou

Funny and true

[epSos-de]

This explains who was watching all of those cat and kitten videos on Youtube.