Edge AI solution trains, runs ML models on MCUs

July 29, 2020 //By Nick Flaherty
The MicroAI Atom engine from One Tech is designed to be embedded on microcontroller units (MCUs) and can now train and run AI models directly at an IoT endpoint.
The MicroAI Atom engine from One Tech is designed to be embedded on microcontroller units (MCUs) and can now train and run edge AI models directly at an IoT endpoint.

One Tech (Dallas, TX) has ported its edge AI unsupervised training technology to microcontrollers to boost analysis in the Internet of Things.

MicroAI Atom is designed to be embedded on microcontroller units (MCUs) and can now train and run AI models directly at the endpoint. This enables silicon manufacturers, original equipment manufacturers (OEMs), smart device manufacturers and smart device owners to reduce the costs of bringing intelligence to the edge and endpoint by at least 80 percent. Enabling unsupervised training on a microcontroller at the nertwork edge allows functions such as predictive maintenance that were previously only available on microprocessors. This differs from the AI accelerators on microcontrollers that run an inference engine that has already been trained.

“This is a groundbreaking phase for the industry. By bringing intelligence to endpoints, sensors and equipment at the network edge, device and equipment manufacturers, along with the owners of these assets, can now have AI-driven intelligence on a low-cost piece of hardware. Training and running a model on an MCU has not been seen before in the industry,” said Yasser Khan, CEO of One Tech.

“AI is shrinking and can run these advanced algorithms. It allows AI and predictive maintenance to move from MPU-based devices to MCU-based devices, with a small footprint and significantly lower price point. Companies in industries such as manufacturing needed this technology yesterday. It is the next evolution of IoT and AI at the network edge.”

Early iterations of IoT solutions primarily consisted of deploying sensors that would pull IoT data points for monitoring assets that the sensors were attached to. This resulted in an influx of data that needed to be further processed and acted upon with edge AI. Now the need for processing IoT data locally to enable automated action is key for IoT deployments, moving from setting static thresholds that would trigger events when the threshold was met or exceeded to creating dynamic thresholds.

“AI models have been largely trained in

Picture: 
Unsupervised edge AI learning on IoT microcontrollers

Vous êtes certain ?

Si vous désactivez les cookies, vous ne pouvez plus naviguer sur le site.

Vous allez être rediriger vers Google.