Scroll Top

Sensors are Entering the New Era of Machine Learning

A new era of IoT sensors with machine learning is here. That was clear at Sensors Converge, where those of us in attendance didn’t talk about the why or the how. Instead, the focus was on the next steps of liability, privacy, and security. We are on the way to “all-out” deployment.

Definitely, we need more data for valuable applications such as environmental monitoring to prevent or at least minimize the scope of the recent Canadian wildfires and the serious air quality problems they caused in the U.S. Or vision processing to simplify machine operation with augmented reality. But at the same time we need to be concerned with how the information we collect will impact privacy. It’s like trying to present an image of the people in the room without violating the rights of any of those people.

We aren’t going to be able to stop the snowballing deployment of IoT devices that are collecting so many different kinds of information, but we need to address the challenges of economics, privacy, security, and the sustainability of our products. All these things and more were discussed at last week’s event.

It won’t be possible to resolve the problem of contextual awareness without defining the hierarchy of POP (point of processing), assigning the different rules and permissions, and implementing distributed computing starting from the sensor itself.

Sensors Converge highlighted sensors for environmental monitoring, pressure and temperature control, level monitoring, humidity, vision for different industries like metal processing, fleet and cruise markets, smart homes, power generation, food and beverage, oil and gas, cold chain, and more. The solutions may solve many problems for CAPEX and OPEX for those industries planned but still raise some common questions.

We see a viable approach in the POLYN neuromorphic front end (NFE) solution, placed right after the sensor, operating to efficiently extract useful data with ultra-low-power requirements and very low latency. These application-specific processor solutions support high resiliency, secures privacy, and provides an easy integration process. Our automation tool – NASP Compiler – could generate a new sensor NFE in very short time-to-market once an original neural network is trained and tested.

A key takeaway from Sensors Converge is that there are plenty of on-sensor solutions now available, and we can say confidently that the sensor will soon be much more intelligent than we have seen before.