Predictable and Explainable AI at the Edge

The project proposed focuses on the serious implications and impact of delivering AI systems in a real world setting. We aim to explore the interplay between complexity and abstraction using a promising all boolean machine learning (ML) method called the Tsetlin Machine (TM). The TM approach will be modified in an effort to improve dependability and verifiability by reducing uncertainty and unrepeatability, and by providing visualisation and analysis tools to extract extended knowledge from the background boolean representation. In addition, we will explore methods to bound the tradeoff between bits assimilated and energy usage,

With this in place, our long term direction is to promote this technology in such a way that those who are thinking more deeply about the pressing safety problems associated with the new agenda in computational autonomy, still have the opportunity to deploy AI based learning systems safely and reliably.