Artificial Intelligence

Artificial intelligence (AI) accompanies us in our everyday life. It often “hides” in everyday applications. AI is seen as the central technology that can make cars drive intelligently and autonomously, make supply chains easier to plan, make production processes more efficient and thus help achieve sustainability goals. It also determines strategies in business and politics and increasingly in administration.

In the computer scientist's toolbox, AI is an important tool. But just as a craftsman doesn't always need the number 10 wrench, the same is true when using AI: Whether AI is a suitable choice for a solution or whether another technical variant is better suited to solve the problem must be decided on a case-by-case basis. “This requires not only expertise in AI, but also in-depth domain knowledge”, says Prof. Dr. Manfred Hauswirth, head of Fraunhofer FOKUS.

Therefore, Fraunhofer FOKUS specializes in the application of AI from practice for practice. AI can be a suitable approach and can replace or complement existing solutions. “The most important prerequisite for the successful application of AI methods is a deep understanding of the subject, the domain in which they are to be applied. Without this understanding, one runs the risk of using AI to solve already well-solved problems, e.g., in control engineering, one more time and possibly worse or with much higher effort”, adds Dr. Tom Ritter, deputy director of Fraunhofer FOKUS. Only with a deep understanding of the domain – with or without AI – an effective, innovative, and efficient solution can be found. In order to be able to judge just this, the work of the FOKUS researchers always begins with a detailed problem and requirement analysis from within the domain. At the same time, extensive methods are available, including AI procedures.

With its many years of experience and specialist knowledge, Fraunhofer FOKUS supports its project partners in solving problems in many domains – be it in the media and entertainment sector, in health care, in public administration or in industry. The research results enable, for example, precise image and object recognition in automated driving, the optimization of processes in public administration, effective quality assurance, or improved prediction of events in the field of public safety. In the field of Smart Mobility, AI is used to optimize processes in the mobility sector.

And not all AI is the same. Rather, the discussion currently revolves mostly around a very specific form of AI, namely pattern recognition through machine learning (ML). ML as a branch of AI is used to find patterns in vast amounts of data to derive correlations that can be used for decision making. In addition to pattern recognition through ML, AI also includes other methods and technologies, such as knowledge-based systems, inference systems, robotics, and intelligent multimodal human-machine interaction.

Research is being conducted on application scenarios for smart cities and predictive maintenance, among other things, and even in the 5G area. The new 5G mobile communications standard will soon shape the communications world. Therefore, Fraunhofer FOKUS uses AI and ML to realize self-organizing 5G networks, e.g., campus networks. These networks can dynamically adapt to usage and security requirements. AI and ML enable an automated and also predictive operation of such networks. There is also a need in the area of video streaming. With ML methods, very large quantities of video material can be analyzed efficiently and quickly with regard to their objective and subjective quality. Based on these insights, recommendations for optimized video encoding can be made. This saves on storage space and transmission costs and improves the video quality.

Fraunhofer FOKUS also works on the standardization, certification, and runtime quality of AI, as well as on IT security by and with AI, amongst others in the field of network security, e.g., for the analysis of data streams at access routers. Our research primarily focuses on integrating machine learning into safety-critical applications and advancing MLOps (Machine Learning Operations) which encompasses the development, deployment, and operation of ML systems. This approach facilitates data-driven innovation by leveraging standardized development processes and automated quality assurance mechanisms. Furthermore, we place particular emphasis on validating ML and generative AI solutions to mitigate risks such as limited robustness, hallucinations, or unethical outputs. AI will also play an essential role in public safety in the future. The main objective is transparency and to enable the authorities to evaluate AI procedures for their potential applications with the expertise of Fraunhofer FOKUS, independent of manufacturers.