SoylentNews is people
SoylentNews
Google Taught an AI That Sorts Cat Photos to Analyze DNA
When Mark DePristo and Ryan Poplin began their work, Google's artificial intelligence did not know anything about genetics. In fact, it was a neural network created for image recognition—as in the neural network that identifies cats and dogs in photos uploaded to Google. It had a lot to learn.
But just eight months later, the neural network received top marks at an FDA contest for accurately identifying mutations in DNA sequences. And in just a year, the AI was outperforming a standard human-coded algorithm called GATK. DePristo and Poplin would know; they were on the team that originally created GATK.
It had taken that team of 10 scientists five years to create GATK. It took Google's AI just one to best it. "It wasn't even clear it was possible to do better," says DePristo. They had thrown every possible idea at GATK. "We built tons of different models. Nothing really moved the needle at all," he says. Then artificial intelligence came along.
This week, Google is releasing the latest version of the technology as DeepVariant. Outside researchers can use DeepVariant and even tinker with its code, which the company has published as open-source software.
DeepVariant, like GATK before it, solves a technical but important problem called "variant calling." When modern sequencers analyze DNA, they don't return one long strand. Rather, they return short snippets maybe 100 letters long that overlap with each other. These snippets are aligned and compared against a reference genome whose sequence is already known. Where the snippets differ with the reference genome, you probably have a real mutation. Where the snippets differ with the reference genome and with each other, you have a problem.
(Score: 2, Informative) by shrewdsheep on Monday December 11 2017, @09:43AM (4 children)
Taykon, thank you for the submission, an interesting read.
Google for the moment, seems to be a one-trick-pony. The problem at hand is related to text data, i.e. DNA sequences, yet the engineers translate it to an imaging problem by creating images that contain the sequences in graphical form (the alignments). Google is very successful in image analysis and it is interesting to see how far you can stretch the approach.
(Score: 0) by Anonymous Coward on Monday December 11 2017, @10:41AM
So they use visualization in order to make the data better consumable to their AI?
Seems their AI is indeed very human-like. ;-)
(Score: 2) by takyon on Monday December 11 2017, @02:53PM (2 children)
Not sure how to take that.
https://en.wikipedia.org/wiki/Alphabet_Inc. [wikipedia.org]
Google (many services like Gmail and Google Docs not listed below)
-DoubleClick
-YouTube
-Blogger
-Android
-Nexus/Pixel (Android hardware)
-ChromeOS/Chromebook
-Chromecast
-Cardboard and Daydream VR
-Google Home (Amazon Echo competitor)
-Google Wifi
Calico [wikipedia.org] (biotech/healthcare/life sciences)
DeepMind
GV (Google Ventures)
CapitalG (another venture capital fund)
X (X Development LLC. [wikipedia.org], formerly Google X)
Google Fiber (dead?)
Nest Labs (smart thermostat and other IoT nonsense)
Jigsaw [wikipedia.org] (a thinktank, formerly Google Ideas)
Sidewalk Labs [wikipedia.org] (urban planning)
Verily [wikipedia.org] (Verily Life Sciences)
Waymo
So we can see that Google is involved in a lot more hardware than it was 5-10 years ago, biotechnology and anti-aging (although being cagey about it), urban planning (presumably data-oriented "smart city" stuff), and driverless cars (which could extract a big payout from Uber before they even launch a single ride for paying passengers). Google X is still doing things like flying Internet/4G balloons over Puerto Rico.
They have not one, but two venture capital funds spreading money around, and many newer Silicon Valley companies have been touched by the Google without being merged or acquired. For example: 23andMe, Cloudera, Impossible Foods, Jet (acquired by Walmart), Medium, Periscope, Slack, Stripe, Uber (the same Uber they are suing), and unfortunate.ly, Juicero. A lot of biotech companies too ("_______ Therapeutics").
A lot of their money comes from advertising, but they also have a lot of tricks up their sleeve, some of which aren't apparent until you look. Maybe their biotech or driverless car subsidiaries will make a lot of money in the future. They are selling more hardware than ever before, and using some hardware internally (particularly TPUs).
I will be surprised if the online/mobile advertising market doesn't shrink massively [digiday.com] in the next few years. Google should be thankful for every penny collected by AdWords/Doubleclick. And they need to grow these side pursuits if they want to survive.
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
(Score: 1) by shrewdsheep on Monday December 11 2017, @08:44PM (1 child)
To clarify: I meant their deep learning efforts. DeepMind is a recent purchase, which indeed does something beside image analysis, however, they also have not changed their basic architecture since AlphaGo. I remember another reinforcement paper from Google about learning to play retro-games (I think Atari). That would be about it.
(Score: 2) by takyon on Monday December 11 2017, @09:24PM
AlphaGo is a bit of a sideshow. Google has "silently" integrated machine learning and TPU hardware into their core products (search, translate, photos, more?):
The Great A.I. Awakening [nytimes.com]
Build and train machine learning models on our new Google Cloud TPUs [www.blog.google]
They got about a decade worth of improvement (at the previous pace) in Google Translate in less than a year.
I expect the Google Home (Amazon Echo/Alexa competitor) is also powered by some TPUs somewhere.
In a way, they are racing against time to improve their machine learning efforts in order to try and use it to save YouTube with censorbots (to prevent advertisers from fleeing the platform in fear of bad press).
[SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]