Apple’s Latest Acquisition Further Illustrates Its Commitment to AI
As we mentioned in our preview of what to expect in the coming year (and decade), some of the big players in tech have made and will continue to make big bets on artificial intelligence being at the forefront of the next wave of innovation. Now Apple, not a company known for being behind the times, is making another significant investment in AI by acquiring Seattle-based AI-on-the-edge outfit Xnor.ai, according to GeekWire. The three-year old startup is a product of the late Paul Allen’s A12 incubator and focuses specifically on image recognition tools that can function on low-power devices that don’t rely on the cloud.
“We make ‘AI everywhere’ a reality by enabling leading global brands to run state-of-the-art deep learning models on anything from a $2 batteryless piece of hardware to the cloud,” according to Ali Farhan, CXO (chief Xnor officer) of Xnor, when Xnor.ai was nominated for an AI innovation award last year. “We empower any developer to deploy countless AI models optimized for the edge with just a single line of code, a future where ‘AI for everyone’ is real.”
Xnor developed an embedded processor solution based on binarized neural networks. BNNs substitute binary values (1s and 0s) for precision values like height and weight, thereby paring down the size of the hardware and cutting down its demand on memory. By avoiding processing in the cloud, image data can be protected from potential privacy concerns while also avoiding high latency delays. Running on a Raspberry Pi Zero, an Xnor demo was able to perform person detection while inferencing at just 8 frames per second. The company also recently demoed deep learning-based person detection on a Lattice ECP5 FPGA (sells for approx. $2) capable of inferencing up to 32 frames per second while consuming only 48mW of power. Its low energy requirements allowed the device to be powered by a tiny solar harvester.
The low-resource functionality of Xnor’s technology make it a good fit for battery-powered devices such as smartphones or tablets as well as closed-network systems that can’t or don’t want to send information to the cloud to be processed, like security cameras and sensor networks. The latter applications fit in with Apple’s recent mantra of preserving data privacy, an issue that’s popped up again for the company recently. On the other hand, the Xnor technology seems to be a bit outside of Apple’s typical modus operandi. It’s very low cost, cloudless operation and pervasive deployment model seem at odds with Apples current focus on high ticket consumer products (phones and computers) and cloud data storage (music and movies). Does this signal a possible shift in focus to include lower cost widely deployed technology or something else? Perhaps a possible Apple watch feature?
The acquisition of Xnor isn’t anything new for Apple, or the rest of the tech industry for that matter. A decade ago, Apple hadn’t bought a single AI-related company -- in 2019 alone, the company purchased over 20 AI-related startups, leading the way among the giant tech conglomerates. According to market research firm CBInsights, venture capitalists invested a record 9.3 billion dollars in AI-related startups, a 72% increase from the year before. There’s a growing sense in the industry that there’s only so much AI-related knowhow available, and the race among the big names in tech to accumulate as much of it as possible has already started. We’ll continue to monitor the AI trends in this space as 2020 progresses.