Powering machine intelligence in mobile and edge devices
Future mobile and edge devices, which are always on and always aware, will require a disruptive solution that delivers the processing power to enable machine intelligence within a very low power profile for applications such as speech and image recognition and heart rate monitoring. At Eta Compute, we have addressed this growing need by developing energy efficient ASICs and novel spiking neural nets-based AI software that can run on this energy efficient hardware. Unlike deep learning models our solution uses sparse representations and operate in an asynchronous mode for both learning and information processing.
This mode of learning precludes the need for many training samples and is more desirable for edge device applications where the amount of resources (both memory and compute) are limited. It also offers a solution for protecting/securing personal information without the need for relying on the cloud for learning. These advantages are beginning to translate into highly efficient learning models for various applications at the edge. A recent benchmark that we achieved was an 8-order of magnitude improvement in model efficiency compared to CNNs and DNNs for key word recognition while consuming just 2 mW of power for inferencing. In this talk, we will provide an overview of this novel framework and our current plans.