Blog: Xnor releases AI2GO, a do-it-yourself software platform for artificial intelligence – GeekWire
Now you too can put a little AI on your device, even if you’re not up on the ins and outs of artificial intelligence.
The way to do it is with AI2GO, a newly released self-serve software platform from Xnor.ai, a Seattle AI startup. AI2GO comes with a set of ready-to-go applications and deep-learning models that can be selected and downloaded with just a few clicks.
Ali Farhadi, Xnor’s co-founder and CXO (Chief Xnor Officer), told GeekWire that the platform is designed for developers and small companies that want to take advantage of AI tools such as face recognition or object classification without having to start from scratch.
“The problem of deploying AI is getting harder and harder, and it shouldn’t be that way,” Farhadi said.
To deploy AI the easy way, AI2GO starts out by laying out an offering of hardware devices that range from a Raspberry Pi circuit board to Linux or Mac OS X processors. (Android, iOS, nVidia TX2 and Windows 10 are “coming soon.”)
Then you choose the kind of application you’re looking for, whether it’s “detecting a person for a dash cam,” or “identifying a pet with a home security camera.” You can tune the application to suit the amount of memory you have to work with, and the amount of latency you can live with.
Once you move the sliders and click the buttons, AI2GO will come up with a downloadable software bundle that includes an inference engine and a pre-trained AI model. “We already have hundreds of models that are good to go,” Farhadi said.
AI2GO’s models are already being used to build solutions for retail analytics, smart-home applications and industrial Internet of Things applications. Farhadi can imagine the platform’s use cases extending out even further, into agriculture, health care, education and language training. Another example would be to create a bundle to use with parking-lot cameras to keep track of how many spaces are open in a given area of the lot.
“There is no other platform close to this one in terms of targeting end users,” Farhadi said.
The concept is in line with Xnor’s mission of delivering “AI everywhere, for everyone.” But AI2GO isn’t necessarily for everyone. Xnor offers more full-featured AI packages and custom-trained models for enterprise-level customers who have more high-powered applications in mind.
Think of AI2Go as a testbed that won’t turn away AI newbies. It gives creative types the opportunity to play around with AI at no cost.
“They can download as many binaries as they want, for as many types of hardware as they want, in order to really understand what’s possible with learning AI at the edge,” said Sophie Lebrecht, Xnor’s senior vice president of strategy and operations.
“What we wanted to do is provide a tool chain where people can download a model, explore, play with it,” she said. “If that person then says, ‘Hey, this model is so powerful, I need to embed it into my commercial product,’ then they would reach out to our enterprise sales team, and we would work with them like we work with a range of customers today, based on a license subscription model. They would essentially pay based on their usage.”
As time goes on, AI2GO will offer a wider selection of models, along with more features such as automatic model training and retraining, plus performance optimization for large-scale development teams.
Xnor was spun out from Seattle’s Allen Institute for Artificial Intelligence a little more than two years ago. Last year it reported a $12 million Series A funding round led by Madrona Venture Group. Earlier this year, the company showed off a solar-powered chip that’s sized just right for edge-based AI applications. But Farhadi said Xnor is still just getting started.
“The mission of ‘AI everywhere for everyone’ is a grand challenge,” he said. “It’s not like anyone can actually achieve that in a year, or two, or three. … AI2GO is a step toward that direction, enabling more and more people to deploy more and more use cases, on more and more devices, under various forms of constraints dictated by their use cases. We don’t think AI2GO is going to fulfill the mission. It’s just a step toward that mission.”