Novel Approach for Machine Learning with Extra Low Energy and Very Light Hardware

The editorial committee of The Japan Society of Applied Physics stated an opinion that “after a few decades, electricity consumption for the AI execution systems would constitute drastically more than half of total worldwide ones, at least if we continue to adopt the von Neumann architecture”. The SNN and neuromorphic architectures would be a reality in the near future with very low power consumption. However, these architectures need to build on the society's infrastructure system. Easily reconfigurable architecture should be realized in the market quickly. This object would be focused on our study. Our architecture called VaCoAl (Vague Coincident Architecture) is a target to eliminate drastically power consumption with light hardware like as commercially available stand-alone equipment as the Raspberry Pi and such. That needs no requirement of the new infrastructure for operation of any AI issue because of its novel search engine architecture that is called LPHV-CAM (Low Power High Volume-Content Addressable Memory). This system concept is structured to search for the extracting vague objects and major decision from the extractions. It just can do additional learning in exactly real-time due to the concept. The extracted items are shown to be distinguished with probability percentage values that makes it possible to perform reasoning.