acm-header
Sign In

Communications of the ACM

ACM TechNews

Google Tests New Approach to Training Machine Learning Models


View as: Print Mobile App Share:
mobile phones

Credit: iStockPhoto.com

Google is testing a new collaborative machine-learning training method in which training data is diffused across millions of individual mobile devices instead of housed in datasets distributed across servers in the cloud. Google says machine-learning models can be trained from user interaction with their Android devices using the Federated Learning strategy.

"Federated Learning enables mobile phones to collaboratively learn a shared prediction model while keeping all the training data on [the] device," say Google's Daniel Ramage and Brendan McMahan. They note this decouples "the ability to do machine learning from the need to store the data in the cloud."

To address the lack of constant mobile device availability for training, Google designed the Federated Averaging algorithm to train machine-learning systems with less communication compared with typical systems. The company also has minimized user disruption by ensuring on-device training only when the device is idle and on a free wireless link.

From eWeek 
View Full Article

 

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


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account