Artificial intelligence startup Neurala Inc., a company that’s built a deep learning neural network used by NASA among other organizations, today announced a major new breakthrough that it claims will finally make the technology useful for consumers.

The company says it has developed deep learning software capable of learning without the need for servers running in the cloud, enabling a new generation of smart cars, children’s toys and industrial machines.

Traditional deep learning systems have largely relied on cloud-based servers to do the computing for them in the background. These systems work, but Neurala said the approach is still inefficient because of what it calls the “catastrophic forgetting” problem that requires machines to re-learn objects they previously knew, whenever they’re required to learn a new object.

“Until now, if an AI system had learned a certain number of objects and needed to learn one more, it would have to be retrained on all of the objects,” the company said. “This traditional method requires using powerful servers that are often located in the cloud.”

But with Neurala’s new technology, learning of incremental objects is now performed on the network edge, eliminating the “catastrophic forgetting” problem once and for all. Neurala didn’t reveal exactly how its done this, but said its method involves combining different neural network architectures in a way that many said was impossible to do.

According to Massimiliano “Max” Versace, chief executive officer of Neurala, this new breakthrough is the “enabler” that companies in multiple industries, from automotive to consumer electronics, have been waiting for.

“The ability to learn on the fly and at the edge means that the Neurala approach enables learning directly on the device, without all the drawbacks of cloud learning. In addition, it eliminates network latency, increases real-time performance, and ensures privacy where needed,” Versace said. “Most importantly, it will unlock the development of a sea of cloudless applications.”

Neurala, which works with NASA’s Jet Propulsion Lab in California, emerged from stealth in January when it raised a $14 million Series A round led by Pelion Ventures. Its Neurala Brain software is described as a deep learning platform that allows artificial intelligence to function similarly to human brain by integrating sight, sound and other senses into one system in a rough emulation of how the brain works. The company claimed the software is also versatile, designed to be installed in driverless cars, drones and robots that need to be able to navigate their surroundings independently, as well as children’s toys and industrial machines.

Explaining how its software works, Neurala said that real human brains may consider the sight and sound of a moving car to estimate its position and place that car in a “mental map” of the world. But it argues that a whole brain AI system such as the one it built is better at performing complex tasks because this native integration enables different senses and modules to complement each other’s deficiencies and shortcomings.

Neurala said it will add the new capability into its Neurala Brain software development kit in the third quarter. In the meantime, the company is demonstrating its technology at Nvidia Corp.’s GPU Technology Conference in San Jose on May 8-11.

Image: geralt/pixabay