acm-header
Sign In

Communications of the ACM

ACM TechNews

MIT Research Could Make Neural Networks Portable


View as: Print Mobile App Share:
The research points to energy savings as high as 73% compared with an unaltered neural network.

Researchers at the Massachusetts Institute of Technology are developing a neural network to work on mobile devices.

Credit: Jose-Luis Olivares/MIT

Massachusetts Institute of Technology (MIT) researchers led by professor Vivienne Sze are developing a neural network designed specifically to work on mobile devices.

Although traditional neural networks require a lot of power, making them unsuitable for mobile applications, the MIT researchers have developed a new approach that pares down the neural network so it can operate efficiently on existing mobile hardware.

The researchers found the mobile neural network can achieve energy savings as high as 73% compared with an unaltered neural network, which would be plausible to run on a phone for a subset of "smart" features.

In order to shrink the neural network down to the proper size, the researchers had to monitor energy usage by building a tool that tracks where energy is being used in the network.

From Extreme Tech
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