Stuttgart, 20 January 2025 – With the international innovation project ‘EdgeAI-Trust’, asvin is setting new standards in the development of secure and trustworthy AI solutions for networked autonomous systems. Edge applications are applications that run directly at the edge of the network or close to data sources such as sensors or IoT devices. Real-time access to central cloud infrastructures can thus be avoided. This is necessary, for example, for applications with high demands on low latency and autonomous functions with limited internet connectivity. Examples include autonomous driving, drones and industrial plants.
EdgeAI-Trust is a project funded by the European Union with around 41 million euros. It brings together 53 partners from 13 countries, including asvin, leading companies such as Infineon, Nvidia and the German Aerospace Center. The aim is to create a robust architecture for artificial intelligence (AI) for edge applications that combines security, reliability, data protection and energy efficiency.
AI at the edge: the next generation of AI
Edge AI researches and develops solutions for the use of AI algorithms directly on local devices and sensors – without resource-intensive dependence on cloud infrastructures. This technology enables real-time processing of data ‘on site’. It is used in safety-critical and process-critical areas such as autonomous driving, industry 4.0 or personalised medicine. Edge AI faces the challenge of functioning reliably in highly dynamic environments while minimising safety risks, for example, due to wrong decisions made by the AI.
asvin’s contribution to the EdgeAI-Trust project
As part of the project, asvin is focusing on developing mechanisms for plausibility checks, for monitoring decisions and for detecting anomalies in AI applications. The project aims to provide a security architecture that takes fault tolerances into account in the context of the application and increases the reliability of the AI-augmented system through corrective mechanisms. In addition, factors such as energy efficiency and cybersecurity must be taken into account in the architecture.
‘With EdgeAI-Trust, we are creating a basis for the development and implementation of trustworthy AI solutions that are not only secure but also meet the sometimes extreme requirements for system security, low latency and energy consumption at the edge,’ explains Mirko Ross, CEO of asvin. ‘Our focus is on helping operational technology (OT) operators to make their AI-augmented edge systems more resilient to wrong decisions and manipulation.’
A European flagship project
The consortium is coordinated by the German company TTTech and includes partners from science and industry. In addition to developing a domain-independent AI trust architecture, a focus is on validation through real-world use cases. The aim is to develop methods and technological solutions that are not only theoretically convincing, but also implementable and scalable in practice.
Relevance for all critical cybersecurity industries
The increasing shift of decision-making through AI to the edge of the network brings new challenges for cybersecurity. In the face of growing threats such as AML or manipulation in IoT systems, one task of EdgeAI-Trust is to provide methods and tools for securing these AI-augmented applications. Attacks using Adversarial Machine Learning (AML), for example, aim to deceive the training of AI through targeted manipulation or to impair the functionality of AI. Attackers exploit weaknesses in the training processes or training data to generate false results in AI applications and thus render the AI models unreliable. In autonomous systems in particular, this can lead to fatal mistakes by the AI, for example in the case of drones, industrial robots or vehicles with life-threatening consequences. But the project also provides important impetus for the legally compliant evaluation and further development of secure AI systems in the context of the EU AI Act, which sets uniform standards for the use of AI for the first time.
If you require background information on AML, data poisoning or adversarial training, I would be happy to arrange an interview with Mirko Ross. Information on other research projects with asvin participation can be found at https://asvin.io/labs/
About asvin
asvin GmbH, based in Stuttgart, offers innovative solutions in the field of cyber security with a focus on IoT, OT and software supply chains. Using technologies such as graph-based cybersecurity knowledge graphs and predictive analytics, asvin strengthens the resilience of critical infrastructures worldwide.
Contact
Konrad Buck
e-mail: k.buck@asvin.io
