The technique was developed by researchers at MIT and the MIT-IBM Watson AI Lab and enables on-device training using less than a quarter of a megabyte of memory.
The need for much less memory is a potential game-changer for deploying AI in a much broader array of devices containing microcontrollers, say the research team.
Microcontrollers – miniature computers that can run simple commands – are the basis for billions of connected devices, from internet-of-things (IoT) devices to sensors in automobiles.
But microcontrollers typically have extremely limited memory and no operating system, making it hard to train artificial intelligence models on “edge devices” that work independently from central computing resources.
As part of the research, the team developed intelligent algorithms that required much less computation to train a model, which made the process faster and more memory efficient.
Their technique can be used to train a machine-learning model on a microcontroller in a matter of minutes, said MIT. Moreever, the technique protects privacy by keeping data on the device.
According to the researchers, training a machine-learning model on an intelligent edge device allows it to adapt to new data and make better predictions. Potential use cases include training a model on a smart keyboard that could enable the keyboard to continually learn from the user’s writing.
“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,” says Song Han, an associate professor in MIT’s department of electrical engineering and computer science, a member of the MIT-IBM Watson AI Lab, and senior author of the paper.