Google AI outperforms developers in writing machine learning algorithms
Options
2stepz_ahead
Guests, Members, Writer, Content Producer Posts: 32,324 ✭✭✭✭✭
https://eandt.theiet.org/content/articles/2017/10/google-ai-outperforms-developers-in-writing-machine-learning-algorithms/
A tool created by Google engineers to automate the writing of machine learning algorithms has outperformed developers in the creation of efficient machine learning systems for some tasks.
The AutoML (automated machine learning) tool was announced earlier this year by Google representatives. Its creation was motivated by the widespread skills shortage associated with coding, particularly for machine learning applications. The Google team built a machine learning system capable of rapidly creating, testing and refining machine learning architectures.
“One way we hope to make AI more accessible is by simplifying the creation of machine learning models called neural networks,” wrote Sundar Pichai, Google CEO, in a blog post.
“Today, designing neural nets is extremely time intensive, and requires an expertise that limits its use to a smaller community of scientists and engineers. That’s why we’ve created an approach called AutoML, showing that it’s possible for neural nets to design neural nets.”
The AutoML approach allows for a “controller neural net” to put forward a suggestion of a basic architecture for a new machine learning algorithm, which can be subsequently trained to perform a highly specialised task. The algorithm’s performance is evaluated and this feedback is used to improve the next attempt.
After repeating these simulations thousands of times, the controller is well-informed about which architectures are likely to perform better at this task. This entire process is automated and takes a matter of hours.
Within just five months, Google has reported that AutoML has begun to generate machine learning algorithms which outperform those written by the developers themselves. An AutoML-generated algorithm proved capable of learning to recognise and categorise images – an extremely complex task – reached a record 82 per cent accuracy.
Even in complex tasks, the AutoML code outperformed code written by developers: in a task in which the system had to mark the location of various objects within an image, the developers’ machine learning software achieved 39 per cent accuracy, while the AutoML software achieved 43 per cent.
Once well-honed enough such that it could be used for practical applications, tools like AutoML could be incorporated into machines of the near future, allowing them to update themselves regularly and even create new programs to solve specific problems.
While machines outperforming humans in highly-skilled jobs remains a fantasy, the success of AutoML demonstrates the slight edge that artificially intelligent systems can have over humans in performing tedious tasks: in this case, the ability of AutoML to rapidly run through thousands of possible architectures would otherwise take months of work for a developer, whose time could be put to better use on more creative tasks.
A tool created by Google engineers to automate the writing of machine learning algorithms has outperformed developers in the creation of efficient machine learning systems for some tasks.
The AutoML (automated machine learning) tool was announced earlier this year by Google representatives. Its creation was motivated by the widespread skills shortage associated with coding, particularly for machine learning applications. The Google team built a machine learning system capable of rapidly creating, testing and refining machine learning architectures.
“One way we hope to make AI more accessible is by simplifying the creation of machine learning models called neural networks,” wrote Sundar Pichai, Google CEO, in a blog post.
“Today, designing neural nets is extremely time intensive, and requires an expertise that limits its use to a smaller community of scientists and engineers. That’s why we’ve created an approach called AutoML, showing that it’s possible for neural nets to design neural nets.”
The AutoML approach allows for a “controller neural net” to put forward a suggestion of a basic architecture for a new machine learning algorithm, which can be subsequently trained to perform a highly specialised task. The algorithm’s performance is evaluated and this feedback is used to improve the next attempt.
After repeating these simulations thousands of times, the controller is well-informed about which architectures are likely to perform better at this task. This entire process is automated and takes a matter of hours.
Within just five months, Google has reported that AutoML has begun to generate machine learning algorithms which outperform those written by the developers themselves. An AutoML-generated algorithm proved capable of learning to recognise and categorise images – an extremely complex task – reached a record 82 per cent accuracy.
Even in complex tasks, the AutoML code outperformed code written by developers: in a task in which the system had to mark the location of various objects within an image, the developers’ machine learning software achieved 39 per cent accuracy, while the AutoML software achieved 43 per cent.
Once well-honed enough such that it could be used for practical applications, tools like AutoML could be incorporated into machines of the near future, allowing them to update themselves regularly and even create new programs to solve specific problems.
While machines outperforming humans in highly-skilled jobs remains a fantasy, the success of AutoML demonstrates the slight edge that artificially intelligent systems can have over humans in performing tedious tasks: in this case, the ability of AutoML to rapidly run through thousands of possible architectures would otherwise take months of work for a developer, whose time could be put to better use on more creative tasks.
Comments
-
They gon engineer themselves.out of a job.
-
No job is safe!
-
-
I read one of the AI projects got shut down because they wrote two different bot that started conversing with each other in a way that the developers couldn't understand so they got shook and cut that ? .
-
The Lonious Monk wrote: »I read one of the AI projects got shut down because they wrote two different bot that started conversing with each other in a way that the developers couldn't understand so they got shook and cut that ? .
They're gonna ? up one day. Tech is advancing beyond even our level of understanding. Man is creating potential monsters. -
where the delete button at?
-
The Lonious Monk wrote: »I read one of the AI projects got shut down because they wrote two different bot that started conversing with each other in a way that the developers couldn't understand so they got shook and cut that ? .
They're gonna ? up one day. Tech is advancing beyond even our level of understanding. Man is creating potential monsters.
Tech is not advancing beyond our "collective" understanding. Also... it's pretty easy to unplug ? . Human error will always be our greatest threat when it comes to technology. -
^^^
oh the ironing. -
If a malevolent A.I is created that can be downloaded to the internet we won't be able to just delete it unless we want to delete the whole ? internet
-
The Lonious Monk wrote: »I read one of the AI projects got shut down because they wrote two different bot that started conversing with each other in a way that the developers couldn't understand so they got shook and cut that ? .
http://www.independent.co.uk/life-style/gadgets-and-tech/news/facebook-artificial-intelligence-ai-chatbot-new-language-research-openai-google-a7869706.html -
The Lonious Monk wrote: »I read one of the AI projects got shut down because they wrote two different bot that started conversing with each other in a way that the developers couldn't understand so they got shook and cut that ? .
They're gonna ? up one day. Tech is advancing beyond even our level of understanding. Man is creating potential monsters.
Tech is not advancing beyond our "collective" understanding. Also... it's pretty easy to unplug ? . Human error will always be our greatest threat when it comes to technology.
you should watch the movie "transcendence"
unpluggin aint gonna do ?
and this is based off replikas makers -
The Lonious Monk wrote: »I read one of the AI projects got shut down because they wrote two different bot that started conversing with each other in a way that the developers couldn't understand so they got shook and cut that ? .
They're gonna ? up one day. Tech is advancing beyond even our level of understanding. Man is creating potential monsters.
Tech is not advancing beyond our "collective" understanding. Also... it's pretty easy to unplug ? . Human error will always be our greatest threat when it comes to technology.
Tech will find a way..
-
The Lonious Monk wrote: »I read one of the AI projects got shut down because they wrote two different bot that started conversing with each other in a way that the developers couldn't understand so they got shook and cut that ? .
They're gonna ? up one day. Tech is advancing beyond even our level of understanding. Man is creating potential monsters.
Tech is not advancing beyond our "collective" understanding. Also... it's pretty easy to unplug ? . Human error will always be our greatest threat when it comes to technology.
Tech is advancing at an accelerated rate that we can't keep up with because of human error not despite it. We're way too smart and curious to not invent something that eventually outsmarts us. Science fiction writers have made their bones off that notion for over 100 yrs. -
We will create our enemy unless we upgrade humans.
-
Bring it on! I'm ready for robot fuckery
-
Bring it on! I'm ready for robot fuckery
I'm not, them mufuckas doing ? I hadn't done in years..
lol at that ? bussin it's ass, tho...
-
-
Speaking of AI looks like Big's infamous phrase "Don't be mad UPS is hiring" is about to become obsolete.
http://mobile.reuters.com/article/amp/idUSKBN1ED1QM