New Research Paves Way for Artificially Intelligent Smartphones

Abhinav Raj, Writer
@uxconnections

Next-gen smartphones may come equipped with state-of-the-art machine learning (ML) models, thanks to a new technique developed jointly by researchers at MIT and the MIT-IBM Watson AI Lab.

When it comes to bringing artificial intelligence (AI) to the devices we use every day, microcontrollers, or rather their limitations, have a role to play. These miniature controllers in charge of simple commands play a pivotal role in many device functions—including controlling touch displays, audio, cameras and myriads of other sensors. 

For long, microcontrollers have governed what’s possible with artificial intelligence on internet-of-things (IoT) devices—including smartphones, iPads, smart home devices and more recently advanced sensors in automobiles. 

Modern microcontrollers, however, suffer from a key flaw—their memory is extremely limited. This places further limitations on what they can do with on-device data. 

Developing a machine learning (ML) model on an IoT device is lengthy, tedious and time-consuming for numerous reasons. For one, training an ML algorithm requires an enormous amount of data and memory—which is why most ML models are developed on servers with larger and more powerful hardware. Since ML models rely on data samples to improve their accuracy, data must be collected from the device and relayed back to servers to train the algorithm—which raises significant data privacy concerns. 

Researchers at the Massachusetts Institute of Technology (MIT) and the MIT-IBM Watson AI Lab have developed a technique that could potentially lead to a revolutionary advance in the incorporation of AI in IoT devices—using microcontrollers.

The team of researchers have pioneered a novel technique that will allow for the development of machine-learning algorithms on microcontrollers—with remarkable memory efficiency. 

Machine Learning (ML) relies on previous data inputs to make future predictions. (Image: Alina Grubnyak on Unsplash)

Developing an approach to enable on-device training of ML algorithms will allow IoT devices to ‘learn’ from new data and make intelligent predictions. For instance, smartphones will continually get better at recognizing the user’s voice, handwriting, accessibility gestures and other functions involving pattern recognition. 

“Our study enables IoT devices to not only perform inference but also continuously update the AI models to newly collected data, paving the way for lifelong on-device learning”, commented Song Han, associate professor at MIT’s Department of Electrical Engineering and Computer Science and the senior author of the study. 

“The low resource utilization makes deep learning more accessible and can have a broader reach, especially for low-power edge devices,” he added. 

The forthcoming generation of artificially intelligent smartphones may be equipped with onboard machine learning algorithms that allow them to learn from user data and get better at performing functions over time. 

This presents fascinating new possibilities for app developers and user experience designers to develop software and features that progressively improve over time as the device learns from its user. 

Would machine learning change the world of smartphones and portable electronic devices? Let us know your thoughts in the comments below.