This deliverable presents the first version of the edge AI algorithms, which serve to the objectives (O 5.1): Identify the key Edge AI hardware, middleware and software design and implementation challenges. Early versions of Edge AI algorithms and methods for resource-constrained devices have been developed to improve distributed learning strategies with continual, hierarchical and incremental learning. On the device level, neuromorphic softwares are developed for neuromorphic hardware. A major focus made is on deploying edge AI run-time inference with adaptation mechanisms in changing environments or conditions, considering multiple inputs (cameras, event cameras) and modalities (vision, audio, EEG) and tasks (2D segmentation, 3D reconstruction). A total of 18 patterns have directly participated in achieving objective 5.1, who have different expertise and collaborations targeted for specific or multiple usecases.
Thales Underwater Systems Thales (Mon,) studied this question.