301.519.9237 exdirector@nesaus.org

8.26.20 – SSN -YARMOUTH, Maine

Tiny chip helps ensure privacy by bringing and keeping data-center-class analytics at the edge

Motivated to find a way to take cloud-level analytics, which are useful to do face recognition, object detection and voice understanding, and put that ability at the edge, Steve Teig, CEO of Perceive, and his team went to work using machine learning and neural networks to build and deploy software, tools and a tiny chip, Ergo, that makes it possible to keep digital analytics at the edge.

“The objective of my effort, Perceive, is trying to upgrade the whole idea of the sensor into something we call a ‘perceiver,’” Teig explained to Security Systems News. “Instead of having a dumb sensor that hands you pixels or hands you audio, but has no idea what it’s looking at or listening too, the idea of putting so much horsepower right at the sensor and the power constraints that you have with an edge device, we can provide data-center-class analytics.”

Doing so reduces the cost of power consumption of sending data to the cloud, but more importantly to Teig, “we preserve the user’s privacy and preserve security by never having the raw data leave the gadget at all.”

Articles of hacking into people’s homes and lives via their smart phones and other smart gadgets abound in today’s media resulting in embarrassment all the way to criminal behavior.

“Amazon employees were sending clips to each other that were recorded from people’s houses [via the Echo] because they though there were funny,” Teig pointed out. “I’m old enough that this kind of stuff really bugs me!”

Responsibilities of manufacturers

Quite frankly, incidents such as these should “bug” everyone. The majority of the world’s inhabitants use technology in some form or fashion on a daily basis and they should feel safe doing so. Since technology is such a prominent part of modern existence, care should be taken at the manufacturing level to give users of said technology confidence in its security.

“First and foremost, [manufacturers should] be aware of it [security] as the first consideration that it actually is,” Teig explained. “I think the cloud companies of the world think that, ‘well, everything’s available for everybody. Hooray!’ [It’s time to] acknowledge there is a concern … that the privacy attribute does matter.”

The second thing is somewhat technical in nature.

“It hasn’t been technically possible — which is what motivated use to start Perceive — to do data center class analytics at the edge,” said Teig. “What people who’ve tried to do stuff at the edge have been forced to do historically is to have pretty low-end computation happening, so the quality of the user’s experience is poor. If it preserves privacy, that’s awesome, but the quality of object recognition is poor because of all the compromises made at the edge.”

After acknowledging privacy matters, manufacturers should invest in a strategy for machine learning, both hardware and software, so as not to compromise the user’s technical experience while still enabling edge capability.

“The possibility that there be recognition of objects, recognition of faces, recognition of voices, in some sense, that’s got to be a good thing,” Teig reflected. “I mean, we recognize each other’s faces; we recognize each other’s voices and part of what enables an intimacy of interaction is that very recognition.”

In Teig’s opinion, it would be great if technology could do all those things as well, with one important caveat: “not sending your data all over the web!” Because of this stipulation, he makes a distinction between face recognition as a technical capability, which he views as a tremendous asset, versus the potential gross misuse of the technology. “Creating a police state … nobody, well at least I don’t’ want that!”

Perceive + ERGO = quality user experience with privacy

At about 7×7 millimeters, or a little over a quarter of an inch, ERGO is about the size of a button on a man’s dress shirt, but it packs a punch at the edge. Because it doesn’t need any external memory, it is ideal for really small, battery-operated gadgets all the way up to larger technologies. What’s more, ERGO is not hackable because the imagery never leaves the chip.

“We’re trying to make this sort of technology ubiquitous, so where ever you have a sensor, you might as well have a perceiver,” said Teig, thinking practically. “What’s the point of having the sensor have to send the data upstream for raw processing?”

As a company, Perceive is selling their ERGO as well as their machine learning applications to end customers.

“Our initial way into the market is security cameras … home security cameras, IoT devices, doorbell cameras and more, with the intention of broadening into appliances and perhaps mobile … drones, robots, toys,” he said.

Teig sees sensors as a historical accident simply based on what technology existed and the timing of that technology’s existence.

“There are 25 billion sensors right now, estimated to be a trillion by 2025,” he educated. “My claim is that it’s not the right thing, technically or societally, that those sensors are all aggregated in the cloud. So, my objective is to build something that’s lean enough, powerful enough and, quite frankly, inexpensive enough that where ever you would have put a sensor — be that in a camera, appliance, toy, doorbell, whatever —  you might as well have something like us right next door and have the whole concept of sensor become something instead.”

Teig and his team at Perceive that putting intelligence at the edge is life-changing for the better.

“People should set the bar and we should do as good a job with these edge gadgets as people have come to expect from the data centers of the world,” he said. “There’s no reason to compromise the quality of your experience in order to maintain your privacy.”