MLOps for Edge AI based Services in Edge Environments

Abstract

We consider MLOps for AI based services that are executing in a DDIL environment. DDIL environments are characterized by denied, disrupted, interruptions and limited network connectivity. We consider the three dimensions along which operations ought to be examined in such environments, and propose an approach to represent different edge scenarios along each of the three dimensions. This results in an architecture for addressing MLOps challenges in a DDIL edge environment, which we describe and illustrate with a few use-cases.