Trending: Review: Amazon’s Twitch NFL live stream shows potential future of live sports online

XNOR.ai computer vision
XNOR.ai’s computer vision tool can recognize objects using software that resides on a smartphone rather than in the cloud. (XNOR.ai Illustration)

“AI for Everyone, Everywhere” may sound like a science-fiction slogan, but it’s actually the name given to software from XNOR.ai that’s already making devices smarter in the real world.

The self-service software development platform is a new product for the Seattle startup, which is also announcing a $12 million Series A funding round led by Madrona Venture Group.

XNOR CEO Ali Farhadi says the new investment will help his company, which was spun out from the Allen Institute for Artificial Intelligence last year, develop a high-end version of the “AI for Everyone, Everywhere” platform for enterprise-level applications.

The name will eventually change, but Farhadi said the phrase captures the platform’s promise. “We’ve been able to scale AI out of the cloud to every device out there,” he told GeekWire.

Ali Farhadi
XNOR CEO Ali Farhadi

The cloud is where artificial intelligence traditionally has taken up residence, due to its need for big-time data processing and GPU-speed processing. But Farhadi argues that the cloud is now turning into a “major roadblocker for AI.”

For one thing, there’s an inherent latency involved in transmitting data back and forth between devices and cloud-based computing systems. For another, there are data privacy concerns that are becoming sharper in the wake of the recent scandals involving Cambridge Analytica and Facebook.

Then there’s the scaling issue: As more devices become woven together into deep-learning systems, thanks to connected cars and the Internet of Things (a.k.a. IoT), Farhadi questions whether even the biggest clouds will be able to keep up.

“Imagine I put a camera at every intersection of every street in America,” Farhadi said. “There’s just no cloud that can handle that bandwidth.”

For all those reasons, Farhadi argues that “the next big step is cloud-free AI,” focusing on a field known as edge computing.

The company is by no means the only company blending AI and edge computing. At this week’s Build conference, for example, Microsoft is highlighting its own initiatives in those realms, including Azure IoT Edge. Other ventures focusing on AI at the edge range from Swim.ai to Intel and Amazon.

XNOR’s key innovation is a computing system that localizes the processing required for AI tasks on the edge, in a device’s CPU. (XNOR.ai’s corporate name refers to the XNOR logic gates that are part of the hardware supporting the system.)

The new self-service platform makes it possible for software developers, even those who aren’t skilled in artificial intelligence, to drop AI-centric code and data libraries into device-centric apps. The platform lets developers specify the memory, latency and power they need, and the code will conform to those requirements.

“We’re making it so simple that any developer should be able to deploy AI into their platform, into their application, into their software, and use it,” Farhadi said.

For example, a developer who’s building an augmented-reality app for a city tour could incorporate object detection, speech recognition and image classification just by checking off a list of parameters. AI tools could be customized for a given app, or supplemented by localized training on the device — for example, to learn how to recognize a specific user’s face or voice.

Farhadi said XNOR is targeting this fall for the release of “an evaluation kit that everybody can play with,” so that developers can see how to make use of the platform for real-world applications.

“We’ll basically charge a fee based on the number of instances deployed,” he said.

He said XNOR has received “an insane amount of interest” in the platform already, based on demonstrations on the industry circuit.

“We have gone beyond that, in the sense that we already have customers … in aerospace, in drones, in automotive, in retail, for photography applications and consumer electronics,” he said. “These are examples of the use cases that are already being used and deployed.”

One of XNOR’s partners is Ambarella, a semiconductor design company that focuses on image processing products for low-power, high-definition video.

“We’re pleased to be working with XNOR to offer their advanced AI capabilities on Ambarella’s low-power, HD camera SoCs [systems on a chip],” Chris Day, Ambarella’s vice president of marketing and business development, said in a news release. “The combination will help to enable a new generation of intelligent IoT cameras, including smart home monitoring devices, capable of delivering analytics at the edge.”

Farhadi declined to name other names. “We don’t have their permission,” he said.

So, how about a smart refrigerator that recognizes the items you put inside, and reminds you when you need to buy more milk? “That’s a very informed guess,” Farhadi said.

XNOR's AI vision
XNOR highlights the differences between cloud-based AI and its own vision for edge-based AI. Click on the image for a larger version. (XNOR.ai Infographic)

Developers can make use of off-the-shelf, ready-to-use training models for object recognition. “We’ve already done that work for them, they don’t need to worry about that,” Farhadi said. But if needed, Xnor.ai can help customers develop specialized data sets that go beyond what’s on the shelf.

The newly announced investments will support XNOR’s growth strategy, which aims to get the self-service platform ready for prime time, scale up enterprise sales and grow the company’s workforce from its curren level of 20 employees to twice that many. To accommodate its bigger staff, and bigger ambitions, XNOR will move into a larger office space in Seattle’s Fremont neighborhood within the next few months.

That’s a fast ramp-up for XNOR, which got its start last year with $2.6 million in seed investment from Madrona Venture Group and the Allen Institute for Artificial Intelligence. “We’ve done really well in our first year, so we’re a cash flow-positive company,” said Farhadi, who founded the company along with chief technology officer Mohammad Rastegari.

Madrona was joined in the newly announced $12 million funding round by Autotech Ventures, a Silicon Valley firm that typically invests in transportation startups; NGP Capital, which has a global fund for IoT investments; and Catapult Ventures.

Madrona managing director Matt McIlwain will continue to serve on XNOR’s board of directors.

“Ali, Mohammad and their team are creating the technologies that will enable the next big innovation in AI  — intelligence and computing at the edge. These edge devices need immediate intelligence to operate efficiently and intelligently,” McIlwain said in today’s news release. “We believe that the next big advances at the edge will come through XNOR’s platform.”

Autotech managing director Alexei Andreev will join the board as an observer. “AI is becoming an integral component in the transportation industry,” he explained. “XNOR‘s lightweight algorithms will enable ‘AI on the edge’ for a broad range of applications, from cars to fleet management to smart infrastructure.”

Update for 10 a.m. PT May 18: This story has been revised to correct the weblink to Catapult Ventures.

Subscribe to GeekWire's Space & Science weekly newsletter

Comments

Job Listings on GeekWork

Find more jobs on GeekWork. Employers, post a job here.