Aug 8, 2018
Qualcomm products mentioned within this post are offered by Qualcomm Technologies, Inc. and/or its subsidiaries.
Chances are, you’re already using devices with artificial intelligence (AI). As AI continues to advance, it’s clear the AI of tomorrow will look different than the AI of today. AI on the edge will help fulfill the technology’s promise and make a difference in how machines learn and process, and therefore the information and experiences they can deliver.
The ability of on-device AI to process data close to the source unlocks low latency, top-notch privacy, superior reliability, and an incredibly efficient use of bandwidth. As Forbes writes, “while we think of AI in the cloud as a huge brain, AI at the edge will be a hive mind of many smaller brains working together in self-replicating and self-organizing ways. AI at the edge will behave as we humans do — each learning from our environment to make locally optimal decisions, on the fly.”
On-device AI is advancing rapidly. In smartphones, for example, its use is currently limited primarily to premium devices, but this won’t be the case for much longer: By 2022, leading IT research firm Gartner predicts 80 percent of smartphones shipped will have on-device AI capabilities, up from 10 percent in 2017.
Of all the ways on-device AI will significantly advance not just smartphones but other sectors like the IoT, automotive, XR, and financial services as well, one of the most fruitful is through biometrics. The response time and bandwidth efficiency advantages of running algorithms locally on the device are important for many AI capabilities, but it’s especially crucial when they’re time sensitive, like so many biometric uses are.
Better biometrics with on-device AI
Much of what people know about biometrics has been limited to fingerprint scanners, but that’s just the beginning. Iris and facial recognition have already broken through, with the more advanced versions using AI to analyze the features of a user’s iris or face, like pupil size or jawline. Soon, on-device AI will enable even more advanced forms of biometrics to become part of our lives.
Gait detection, or identifying a person by their walk, for example, has been researched for decades without much advancement — until now. Recent advancements in accuracy made possible by AI have turned gait detection into something viable. Earlier this year, researchers at The University of Manchester attained an accuracy of 99.3 percent, according to a paper published in the Transactions on Pattern Analysis and Machine Intelligence (TPAMI) journal. The system analyzes individuals’ steps using floor sensors, and with AI, getting that last percent of accuracy is often the most challenging part. With more progress in on-device AI, it could eventually be deployed in places like airport security and even used to diagnosis medical conditions.
Voice is another area primed for AI-powered biometrics. Consumers are already using voice authentication and voice assistants, but progress in on-device AI will be crucial in developing voiceprint biometric technology that closely resembles natural language. People are accustomed to quick responses and a seamless back-and-forth in their conversations. Consequently, a natural voice interface can only tolerate so much latency before the user experience suffers. With the reliability and latency improvements on-device AI offers, our machines will finally be able to sound more like us and less like robots.
Empowering industries through on-device AI and biometrics
Across sectors, these new and improved use cases that rely on biometrics are changing the devices we’re already using while paving the way for new technology.
In the IoT world, the evolution from ordinary devices (like light bulbs and thermostats) to those that are more advanced (like smart assistants and appliances) has largely relied on on-device AI. The data collected from sensors, microphones, and cameras in these more technologically advanced devices creates the opportunity for the training and predictability capabilities of machine learning. And the best place this can occur is often on the device itself, for reasons ranging from privacy to lower latency and higher reliability. Take a smart home security camera, for example. The ability of its facial tracking application to analyze a video feed on the camera locally allows it to be much more responsive, since there’s no waiting to send it back and forth between the network edge and the cloud for processing.
Automotive is another industry that is expected to be greatly impacted. Inside the car, natural user-interfaces, personalization, and driver awareness monitoring can begin to be powered by AI on the edge. This will allow for personalized experiences, such as playing your favorite music or adjusting driving dynamics, based on biometrics and preferences. And outside, AI on the edge can pave the path to autonomy with surround view perception, path planning, and decision making.
Qualcomm Technologies leading the way
Qualcomm Technologies has a history of leading the way in biometrics. The launch of Qualcomm Snapdragon Sense ID 3D Fingerprint Technology in 2015 was the mobile industry’s first 3D fingerprint authentication technology based on ultrasonic technology. We then launched a new and improved Qualcomm Fingerprint Sensors platform. This suite of features boasts 6X more accurate biometric performance than its predecessor and has sensors for display, glass, and metal, as well as detection of directional gestures, underwater fingerprint match, and device wake-up. What’s more, it’s the first commercially announced integrated ultrasonic-based mobile solution to detect heartbeat and blood flow for an improved mobile authentication experience.
We’ve also introduced biometrics to Extended Reality (XR) with the Qualcomm Adreno 630 Visual Processing Subsystem. The Adreno Foveation feature — which is also included in the Snapdragon 845 Mobile Platform — uses eye-tracking to see which parts of a given screen a user is looking at, then focusing bandwidth and power on rendering images in those areas. For the user, this is engineered to maximize memory efficiency and battery life for incredibly long tetherless experiences.
We’ve seen some of the magic AI and the cloud can produce in a field like biometrics. By moving some of that processing on device to take advantage of lower latency, higher reliability and bandwidth efficiency, and greater privacy that on- device processing supports, we’re leading the way to a whole new class of biometrics — one that will shape industries and transform lives.