Blog Post

How to Shrink the Apple Watch for Your Finger

Embedded Devices, Machine LearningOura has mastered the art of embedded edge computing, a new field of edge machine learning that gives silicon a personality.

Adam Benzion

January 3, 2022

When I woke up this morning after a rare good night's sleep, the first impulse I had was to sync my Oura Ring with my phone. I wanted to learn about the conditions that led me to one of life’s most rewarding and healthy pleasures: a deep, restorative sleep. I’ve been using my Oura Ring for over six months and I rarely think about battery life, notifications or missed calls. Instead, I better understand my sleep, workouts, and caloric burns with near 100% sensor accuracy. I wear my Oura Ring on my finger and it occupies such little space that I forget it’s there (unlike my Apple Watch that always felt intrusive on my wrist). I optimized for insights, form factor, and simplicity versus another computer I need to manage. Instead, the Oura Ring takes care of me, runs for six days on a tiny battery, and sports purposeful micronized components that seem too good to be true. It’s everything I had hoped the Apple Watch would deliver, except it fits on my finger. 

Read the Giving hardware a neural soul sectionGiving hardware a neural soul

Oura has mastered the art of embedded edge computing, a new field of edge machine learning (ML) that gives silicon a personality. The company is making innovative and efficient use of data harvested daily from tiny infrared sensors, body temperature sensors, oxygen sensors, accelerometer, and a gyroscope to help you better understand your physical wellbeing. This device knows more about my vitals than my doctor. To evaluate sleep patterns, for example, instead of going into an uncomfortable medical facility and sticking electrodes to my face, the Oura Ring is capable of correlation, learning, and optimizing data, operating as a companion brain without needing high-end medical equipment. The Oura Ring can ingest data, use ML models, analyze anomalies, learn, and update data, all to provide me with daily personal insights that I’d never even expect from a bulky Fitbit or the Apple Watch. 

Oura uses tinyML, a specific branch of ML found in small edge devices that have little memory or power, to measure deviations from the norm, to check biometrics, heart rate, goal progress, sleep readiness, and even give you a sleep score. This new ability to ingest and model data from tiny devices is finding its way into hardware that ranges from the likes of NASA, Microsoft's Percept, and Apple’s AirPods. 

“Building machine learning models for hardware that has little memory, no Internet connectivity and little power, without sacrificing end-user experience, requires specialized compression and intelligent model design capabilities,” explains Jan Jongboom, CTO and co-founder of Edge Impulse. “You can use traditional tools for embedded ML that take an enormous amount of time, skill and costly engineering hours to build, but next-generation developers and enterprise product designers don’t have the time, resources and patience for that. Looking forward, it’s about ‘getting it done’ faster, better and at a lesser cost than before,” he says, “and that’s what modern ML technologies are all about.” 

Read the IoT’s new best friend  sectionIoT’s new best friend 

The connected devices that make IoT electronics tick seem archaic and out-of-touch without ML. In the world of IoT, users get to use rich telemetry but with almost no insight into intelligent, autonomous decisions. This new generation of IoT + ML brings AI into our pockets and business machines by shrinking deep learning networks to run thinking models on every kind of hardware, big or small. With this new capability, ML-powered IoT devices could run at less than one milliwatt, making use of scarce resources so efficiently, a device can run on a single battery charge for years, connecting by radios only when needed, temporarily, to convey critical “wake” detection of anomalies, events, sounds, motions, and images. The model determines how devices run when they are on or off and what data is needed to achieve a goal. This new technology will proliferate AI-powered IoT devices, giving IoT the magic it always wanted, and scale to billions of inexpensive and independent sensors. The ability to provide brains to devices everywhere will impact almost every industry: retail, healthcare, transportation, wellness, agriculture, fitness, and manufacturing.

While the Oura Ring is my favorite poster child of edge ML, I came across many other applications using this technology in recent years in ways that I’ve never seen before. In conservation, Open Acoustic Devices created a low-cost, full-spectrum acoustic logger based on a processor from Silicon Labs that can listen to audible frequencies up to 384,000 samples per second. Smart Parks, a Dutch-based wildlife conservation organization, is deploying elephant collars called ElephantEdge that run specific ML applications for park rangers in Botswana and Mozambique, with over eight years of battery life per every collar, unlike anything we’ve ever seen before. “As edge ML technology advances, we can make use of less powerful accelerometers, microcontrollers, and batteries, and still make the devices longer-lasting, smarter, and more useful,” said Tim van Dam, founder of Smart Parks. “I think the most important thing to understand about this technology is that we haven’t even begun to scratch the surface of what we can achieve." 

AudioMoth Vault

While cloud computing runs almost everything we use today, edge ML is its new unicorn friend riding a rainbow spectrum. Looking into the future, where there will be faster networks available everywhere and edge ML will enable humans and machines to interact symbiotically; everyday technology will no longer intrude into our lives, but instead it will integrate, complement, and enhance life in all sorts of tiny ways. 


Are you interested in bringing machine learning intelligence to your devices? We're happy to help.

Subscribe to our newsletter