Proposals need to particularly address several but not necessarily all following aspects:

• Development of end-to-end AI architectures including the continuum of AI-based algorithms, devices, as well as interoperability, upgradability and trusted exchangeability through standardized APIs across resource-constrained deviceconnected systems (including interfaces between sensing/actuation and computation), edge processing units (to facilitate tight combination of logic and memory providing wide interface lanes and low latency), on-premises servers, etc.
• Development of the next generation of HW/SW edge AI technologies that support fully collaborative AI by allowing devices to learn at the edge and to incrementally evolve their AI/ML models over time (supporting e.g., transfer learning and meta learning), as well as to coordinate with one another (orchestration), to combine their updates and to send the result out (potentially including cognitive reasoning tasks) in a meaningful manner.
• Development of design tool chains featuring automated tasks (e.g. AutoML) and standardized interfaces to deliver optimized edge AI solutions in heterogeneous embedded systems. The tools chains should support fast and early exploration and efficient mapping of complex AI structures such as Deep Neural Networks (DNNs) on embedded architectures and implementation of algorithms for selection, compression, handling and collection of meaningful data sets, in order to generate efficient collaborative AI applications at a rapidly accelerated pace.
• Development of assurance frameworks for edge AI trustworthiness8 with a primary focus on privacy, robustness, safety, security as well as evaluation of associated error/uncertainty bounds, certification/qualification management, and paths towards explainable AI.
• Establishing sustainable impact and creation of an industry-driven community by building tightly integrated open edge AI platforms and ecosystems, with support for the entire lifecycle of edge AI-based solutions, including design toolkits supported on public and royalty-free standards. These ecosystems must promote trusted and transparent governance frameworks for the whole digital infrastructure stack, applications and users/stakeholders.
•  Development of benchmarking frameworks with meaningful metric sets of relevance to edge AI, with particular focus on at-scale technology availability, energy efficiency, security, safety and privacy for a given set of specifications and constraints. A special care on measuring the environmental impact of the developed solutions, for example using Life Cycle Assessment (LCA), must be considered to meet the sustainability challenge.
• Support of open-source is strongly encouraged to make the platforms (open-source hardware, software, training datasets, etc.) accessible and to spread it easily in the edge AI community and in the market.