Massachusetts Institute of Technology (MIT) researchers have come up with a machine learning (ML) method to reduce the amount of memory required for Internet of Things (IoT) devices.
The researchers boosted the efficiency of TinyML software by analyzing memory use on microcontrollers running convolutional neural networks; they applied a new inference technique and neural architecture to address imbalanced memory utilization-induced bottlenecks, reducing peak memory usage four- to eight-fold.
When deployed on the next-generation MCUNetV2 tinyML vision system, the method was more accurate than other ML techniques running on microcontrollers.
"Without [graphics processing units] or any specialized hardware, our technique is so tiny it can run on these small cheap IoT devices and perform real-world applications like these visual wake words, face mask detection, and person detection," said MIT's Song Han. "This opens the door for a brand-new way of doing tiny AI and mobile vision."
From MIT News
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA