Key technology

Focus: Safety for AI

Competition for secure machine learning enters phase 2

Dr Daniel Gille, Head of the project and Head of the Artificial Intelligence Unit at the Cyberagentur
Dr Daniel Gille, Head of the project and Head of the Artificial Intelligence Unit at the Cyberagentur

The Agentur für Innovation in der Cybersicherheit GmbH (Cyberagentur) has launched the second phase of its “Robust and Secure Machine Learning” research competition. Following a careful evaluation of the eleven short concepts submitted, in which members of the jury from the BSI and the German Armed Forces also took part, five selected teams will now continue their research. Their focus is on the security and robustness of neural AI systems against targeted attacks and unforeseen inputs.

The first phase of the competition, which began in early April 2023, aimed to develop innovative concepts for robust and secure machine learning (ML). The call for projects covered five key areas of research: data validation, model verification, system embedding, development of hybrid models from neural and symbolic components and end-to-end verification. The multi-phase project is intended to drive forward various innovation approaches over the five-year term.

The five concepts of the consortia that were successful in the first phase of the competition are characterised by diverse and exciting approaches with a high degree of modularity and multimodality. In phase 2, the consortia will enter the long concept phase. This will last six months. Three consortia will then be selected to continue basic research and the development of prototypes.

“Overall, we received a wide range of research approaches – from software frameworks and toolkits to more hardware-centric and algorithmic concepts,” says Dr Daniel Gille, Head of the project and Head of the Artificial Intelligence Unit at the Cyberagentur, describing the results of the competition. “For the evaluation process, we therefore worked with our jury to closely examine which proposals could be recognised as potentially ground-breaking approaches to increasing the robustness and security of various AI methods in the areas of internal and external security.”

Pre-Commercial Procurement (PCP) was used for the “Competition of Ideas”. This specific procurement procedure for services in research and innovation guarantees competition-based and transparent research in several phases and risk-benefit sharing. For the evaluation, the Cyberagentur has formed an expert jury made up of its own researchers and two external members. A productive partnership was established with representatives of the Federal Office for Information Security (BSI) and the Bundeswehr Digitalisation and Capability Development Centre for Cyber and Information Space (ZDigBw), which will continue in future phases. “This guarantees that future users of the research results are already involved in the competition process and will support the research work in a goal-oriented manner with their well-founded and application-oriented expertise,” says Dr Gille, explaining the composition of the jury.

Further information: https://www.cyberagentur.de/bevor-maschinen-ueberlistet-werden/

Newsletter

Your update on research, awarding and co.

Subscribe to our scientific newsletter. In this way, you can find out promptly which research projects we are currently awarding, when partnering events, symposia or ideas competitions are coming up and what’s new in research.