Computing capability continues to get cheaper and more powerful every year even as Gordon Moore’s eponymous law, forever seemingly immutable, has come under threat. Computers are so common as to be nearly invisible. Embedded computers help cameras intelligently monitor stock levels in stores and warehouses, density sensors measure foot traffic, and increasing numbers of sensors power industry and manufacturing. In the home; wearables, appliances, toys, and smart home devices contain computers within.
Cloud computing has driven unprecedented growth, but it can’t always get the most out of such embedded computers. The cloud paradigm suggests that inputs for meaningful computations be sent over a network to virtualized servers, where the computations are performed and the outputs sent back or acted upon. The approach carries compromises. It requires internet connectivity, which can come at the expense of power consumption and requires a reliable high bandwidth connection. It introduces variable latency, frustrating applications that require real-time processing. It engenders costs in processing large volumes of data, often on third-party servers, which can mean more exposure to security and privacy breaches. As a result, many embedded computers hold untapped potential. The sensors on the engines of a Boeing 787 generate a terabyte of data every 24 hours, much of which never see the light of day.
Machine learning at the edge provides a solution. If complex computations can happen on the embedded computer, instead of being sent back to a cloud server, a new world of applications opens up. Of the vast amounts of data exhaust generated by today’s embedded computing devices, machine learning models can infer relevant or actionable data. Appropriate actions can then be taken automatically — a temperature adjustment by a thermostat, for instance. Alternatively, just that relevant data can be sent back to the cloud, such as a snapshot of an alert from a security camera or a worrisome sensor reading. This approach drastically reduces power consumption, can happen in real-time, and requires little to no network connectivity.
The microcontrollers within such embedded computers can thus enable new applications in mobility, industry, retail, consumer devices, and agriculture. But the large and passionate communities of developers building for these devices haven’t previously had access to an end to end developer-first platform enabling them to create, train, deploy, and manage models efficient enough to perform on embedded devices.
Enter Edge Impulse. The product allows developers building applications for embedded computers to manage data sets as well as design, train, and test sophisticated models. Once designed, efficient models that can run at the edge are easily deployed to a wide and increasing variety of microcontrollers. Intuitive training capabilities (as easy as linking a smartphone’s sensors via QR code, or as complex as ingesting the exhaust from a variety of supported edge devices) support a vast array of potential data types: audio, bio-signals, device logs, accelerometers, radar, location, and images. The team continues to introduce functionality. New templates and datasets, an expanded selection of algorithms and supported hardware, and functionality making it easier to collaborate and manage models in production will continue to empower developers and teams building embedded applications, who have already built thousands of projects on the platform since it launched earlier this year.
True to the team’s developer-first mindset, Edge Impulse works closely with open source communities, is home to a growing developer community, is partner to organizations such as Hackster (where hundreds of thousands of passionate embedded computing developers congregate) and hardware providers including Eta Compute, Arduino, and ST Electronics. I encourage anyone interested to look through the documentation, which makes getting started extremely easy and requires nothing more than a smartphone. And if the mission resonates, the company happens to be hiring.
Edge Impulse founders Zach Shelby and Jan Jongboom live and breathe embedded technology. They put developers first and work tirelessly to bring their vision of edge computing to life. The team they are building is similarly ambitious and obsessed. We’re proud to partner with the team in their recent round of funding.
Asad Khaliq, Mark Kraynak, and the Acrew Capital Team