Artificial Intelligence firm Movidius has announced a USB stick that can be plugged into existing systems to add deep learning functionality that can allow them to understand the world around them or recognize objects. It’s called Fathom and will allow powerful neural networks to be moved out of the cloud, and deployed natively in end-user devices.
It has a built in ultra-low power, high performance Myriad 2 processor inside and can run fully-trained neural networks at under 1 Watt of power. With a standard USB connectivity, it can be connected to a range of devices. After plugging it into a USB 3.0 port developers can immediately start running a fully-trained neural network while consuming very little power. Fathom could give a robot the ability to see and understand objects and navigate around them. It might also let them recognize human faces and expressions.
Neural Networks are used in many applications such as object recognition, natural speech understanding, and autonomous navigation for cars. Rather than engineers programming explicit rules for machines to follow, large amounts of data are processed offline in self-teaching systems that generate their own rulesets. Neural networks significantly outperform traditional approaches in tasks such as language comprehension, image recognition and pattern detection.
It’s possible to add deep learning to existing systems using cloud technologies, but that ties you to an online service. The less you have to feed into the internet the better.