Fritz lands $2M to help bring neural networks to mobile

Mobile developers can use Fritz to process live video, images, etc. on the edge

Face detection, object identification, voice recognition. These are technologies using neural networks (algorithmic constructs that allow software to learn) that enable us to deposit a check, translate in real-time, and apply a doggy filter to your image.  

Until recently, neural networks were stored on the server and weighed a ton of bytes. Processing through neural networks required moving data from the edge device – smartphone, tablet, smart home, car, etc. – to the cloud, then transferring it back to the gadget - a process with many shortcomings.

A new startup wants to help shorten that process. Fritz Labs is a platform that helps take ML and AI algorithms out of the cloud and into the edge. The startup announced this week that it landed a $2 million seed round.

“You want to use a neural network to process live video, say, from a smartphone – there is just not enough time or bandwidth to send live video up to the cloud, process it and send it back down to your phone again in real-time," said Jameson Toole, CEO of Fritz.

The new way to process data, in a somewhat revolutionary shift, is to avoid the loops: do it on the gadget.

“Your phone collects all this data and processes it directly,” Toole said. “This is a lot faster, a lot cheaper for software developers: they don’t have to pay these really high fees for AWS [Amazon Web Service] or Google Cloud. It’s also more private, because your data doesn’t ever leave your phone, it stays there.”

Last summer, as Apple declared its Core ML implementation, Toole and Dan Abdinoor decided they would shorten the space between the developer and the edge device.

The two founders are both engineers, with years of experience working with data science and ML models. Abdinoor held leading engineering positions at HubSpot; Toole interned at Google, then got his PhD from the Massachusetts Institute of Technology. The two met when working at Jana, a mobile advertising company, and decided to start something on their own.

“There were always rumors about the special processors, the neural engine that Apple was going to put in the new iPhone X, so we realized this machine learning technology was going to be absolutely huge for developers," said Toole. 

The team set out to create the infrastructure and a set of tools to make it easier. In April, Fritz launched in Boston.

In September, Toole said, they closed a $2 million seed round with Hack VC and Uncork Capital, among others, following the lead of Eniac Ventures. Vic Singh, Founding General Partner at Eniac Ventures in the company’s press release called moving from cloud computing to edge devices “a natural progression.”

Thanks to the new funding, this week Fritz was able to offer free early access for developers building apps with TensorFlow and Core ML.

Currently, the company’s main page prompts visitors to sign up, in a chance to try out the product. Later, paid premium features and services will be added.

Among offered features are data monitoring and analysis, Toole said, to ensure a neural network model runs quickly and delivers smooth user experience across all devices. Unlike Core ML, Fritz helps developers get into production, deploy and update the models.

A forthcoming premium product is optimization.

“Machine learning models can be quite large,” said Toole. “For example, an image recognition model might be 500MB, and you would never, as a developer, want to ask a user to download an app that is 500MB. We have tools that can shrink these models down to something much lighter and easier to put into an app.”

Fritz is at an intersection of two big trends, Toole said. The first is the advance in ML and AI, with deep learning and neural networks. Computers learned to do things that developers want to use in their products and services. The second tailwind is the explosion in the number of sensor devices like Amazon Echo, Google Home, and self-driving cars.

“There is so much data collected from cameras, microphones, accelerometers on your edge devices that we’re reaching a point where we can’t feasibly send all that data back to the cloud for processing. You are forced to move algorithms from the cloud to the edge.”

Said Toole, “We’re really excited to announce the company and our early access program.”

Image: Nextbillion.net