EdgeAI-Trust EU Project
EdgeAI-Trust hat zum Ziel, die künstliche Intelligenz (KI) so weiterzuentwickeln, dass das Vertrauen der Öffentlichkeit in die KI gesteigert werden kann. Entscheide der KI sollen erklärbar und nachvollziehbar werden. Der Hauptfokus liegt auf KI-Systemen, die ohne Cloud-Unterstützung funktionieren, also "on the edge".
Beschreibung
Develop a trustworthy domain-independent AI architecture
The architecture will support the development of collaborative AI-based systems with orchestration, upgradeability, manageability, reliability, real time, safety, security, and energy efficiency. The architecture will support a continuum of heterogeneous AI-based algorithms and devices along standard APIs for interoperability and trusted exchange. This will range from sensor-actuation, large scale device-connected systems, edge processing units to the cloud. While the architecture is domain-independent, it will enable domainspecific instantiations for trustworthy collaborative AI systems.
Create reliable and collaborative large-scale edge AI solutions
Develop the next generation decentralized HW/SW edge AI technologies, with a particular focus on safety-critical and security related systems. The technologies will support fully collaborative AI by allowing heterogeneous devices to learn, adapt at the edge to cognitive reasoning tasks. Decentralisation of resources and processes among different entities/devices enable dynamic reconfiguration of processes in a resource constrained environment.
Increase trustworthiness of edge AI solutions
Considering the upcoming EU AI Act, here the focus is on explainability, reliability, safety, security and robustness of edge AI solutions. We will achieve that by focusing on
- rigorous plausibility checks,
- monitoring of AI decisions and anomalies,
- explanation of situational awareness, and
- also include reliability and fault tolerance in a dynamic zero trust environment.
Optimization and validation of decentralized edge AI solutions
The goal is to provide methodology and tools that enable the optimization and validation of AI systems based on the EdgeAI‐Trust architecture and solutions from the development platforms to the finished product. Automation supports developers by taking care of the constraints (like accuracy, latency, resource constraints and reliability) due to the limitations of the edge AI systems. By that, rapid improvement and deployment of Edge AI‐based systems is provided.
Eckdaten
Projektleitung
Co-Projektleitung
Projektteam
Stephan Bachmann, Etienne Gubler, Dr. Amin Mazloumian, Simon Meli, Janosch Morf, Johannes Neyer, Simon Elias Schneider
Projektpartner
Miromico AG; Uyan Electronics GmbH
Projektstatus
laufend, gestartet 06/2024
Institut/Zentrum
Institute of Embedded Systems (InES)
Drittmittelgeber
Horizon Europe