Sign In

Communications of the ACM

ACM TechNews

Google Researchers Are Teaching Their AI to Build Its Own, More Powerful AI

View as: Print Mobile App Share:
Robots coding.

Google has devised a way to use neural networks to build better neural networks.


Google has unveiled a new approach to machine learning in which neural networks are used to build better neural networks, in essence teaching artificial intelligence to teach itself.

The new technology, AutoML, can develop networks that are more powerful, efficient, and easy to use.

"The way it works is we take a set of candidate neural nets, think of these as little baby neural nets, and we actually use a neural net to iterate through them until we arrive at the best neural net," says Google CEO Sundar Pichai.

The reinforcement-learning process enables computers to tie trial and error to some kind of reward.

The process normally takes a massive amount of computational power, but Google's hardware is approaching the point at which one neural network can analyze another.

Tests of the system indicate AutoML might be smarter at identifying the best approaches to solving a problem than human experts.

From ScienceAlert
View Full Article


Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA


No entries found