MARVEL will deliver an AI-based framework that will provide multimodal perception of audio-visual scenes and will employ advanced Machine Learning model distribution and decision offloading mechanisms. These techniques will enable fast reaction to data streams and will help decision makers gain situational awareness in short times. The WP3 objectives are to

• define and execute measures ensuring privacy preservation during data processing.
• follow a personalized federated learning approach to train and execute ML models at the Edge, Fog and Cloud.
• train new or updated ML algorithms for audio-visual classification and analytics.
• optimise ML deployment in the E2F2C infrastructure in a continuous manner.
• deploy ML algorithms at all layers of E2F2C.
• implement light-weight ML models for deployment at the edge.