MARVEL - D3.2: Efficient deployment of AI-optimised ML/DL models – initial version

The purpose of this deliverable is to describe the MARVEL Edge-to-Fog-Cloud framework as the deployment layer for the AI/DL MARVEL components. This framework incorporates the deployment logic that is hidden behind MARVdash, the proposed Kubernetes dashboard for instantiating services as orchestrated containers, and deploying them to desired execution sites based on optimisation strategy. The main goal of the optimisation strategy is for MARVEL components to be deployed into Kubernetes nodes based on their resource requirements and the resource offerings of the actual nodes. Moreover, this deliverable will describe methods for compressing machine learning algorithms/models based on the resources available at the edge (e.g., reducing the size and operation time of million-parameter deep learning models). Such compression could minimise the computational overhead on the edge servers.