Do you know how an autonomous car can differentiate a person from an icon in traffic? How does the music streaming app, Spotify, manage to indicate a song for you? Why is Gmail's spam filter so effective? The explanation for this is a type of artificial intelligence (AI) called deep neural networks. These systems are excellent for performing data recognition and classification tasks, however, they need processing power and memory to be executed in real time (something difficult to achieve in a common smartphone).
Researchers at Northeastern University were able to run deep neural networks on a smartphone without internet
Recently, researchers at Northeastern University were able to create a way to run deep neural networks on a smartphone or similar device. Using their method, networks are able to perform tasks 56 times faster than they had previously been able to do, without losing accuracy.
You can hold the power of #deeplearning in your hand (literally) with help from a @NortheasternCOE engineer. @RealAAAI https://t.co/5H6lZ2gmea
– Northeastern U. (@Northeastern) January 30, 2020
The assistant professor of electrical and computer engineering at the university says:
"It is difficult for people to be able to run neural networks in real time on a smartphone or mobile device. But we can make most deep learning applications work in real time."
What is more often than not the need for the mobile device to be connected to the internet to have access to a deep neural network. In this way, the cell phone collects the data, but all processing is done through remote servers, which is why you are unable to give commands to the Siri voice assistant when the iPhone is in airplane mode.
Yanzhi and his teammates were able to find a way to reduce the size of the neural network model and automatically produce a code to run it with greater efficiency. This project can allow deep neural networks to be implemented in smartphones ready for use without the need for internet access. Yanzhi says:
"There are so many things that need intelligence. Medical devices, wearable devices, sensors, smart cameras. All of this, they need something that improves recognition, segmentation, tracking, surveillance and many other things, but these are currently limited. "
What did you think of this innovation? Comment below and share your opinion with us!
Source: northeastern news, techxplore
We recommend reading: