Together with Prof. Dr. Martin Schell, Prof. Dr. Thomas Wiegand heads the Fraunhofer Heinrich-Hertz-Institut (HHI), which has been working very successfully for years on research into communication and data processing methods. Since 2018, Prof. Wiegand has also headed the ITU/WHO focus group "Artificial Intelligence for Health", which evaluates AI algorithms in health care worldwide.
The Fraunhofer HHI would like to help shape the digital society of tomorrow with excellent research and development work and is accordingly broadly positioned. As Institute Director, you have a good overview: which aspects of artificial intelligence are being researched in your Institute?
Our research in artificial intelligence ranges from the basics to applications. Together with Prof. Klaus-Robert Müller from the TU Berlin, we have, for example, made great strides in basic research in the area of the explainability of AI. In doing so, we can show what this result has achieved with a classification result which was delivered by a deep neural network. For this purpose we invert the non-linear network and show which of the mechanisms in the network and which part of the input signal are relevant. Our work in this area has met with great international approval, which is reflected in Best Paper Awards, numerous international publications, including the recently published Nature Communications article, and many inquiries from industry. Furthermore, in the field of basic research we also deal with compression, distributed machine learning and the development of novel methods of machine learning for modern mobile radio networks such as 5G.
"Cooperation" is an important keyword for the work of the HHI as an interface between research and industry. How are you networked, and which joint projects exist beyond the Institute's boundaries?
For many application projects, cooperation with researchers from the relevant fields, such as medicine or the automotive industry, is of utmost importance. At the Fraunhofer HHI, Dr. Wojciech Samek and Prof. Slawomir Stanczak and their respective teams are actively involved in successful research collaborations in machine learning.
Dr. Samek and his team are developing novel methods for the analysis of medical data. These include time series data such as ECG, EEG, EMG, image data such as CT, MRI, fMRI and data from mass spectrometers. The analysis of health data has various challenges, in particular due to its high dimensionality, its strong spatial-temporal correlation, lack of measurements and the comparatively often small sample size. The research focuses on deep learning-based methods which address these challenges.
Research on 5G network technology and its possibilities is another major hobbyhorse at the HHI.
Prof. Stanczak and his team are involved in the design and optimisation of modern radio access networks, with a clear current focus on 5G. New 5G applications such as tactile networked driving or reliable remote machine operation place extremely high demands on the quality of service of 5G networks. Thanks to machine learning, critical states which can result in extreme technical faults are detected at an early stage in order to minimize or completely avoid their effects. Such pro-active mechanisms form the basis of reliable and resilient radio communication in the automotive and industrial sectors. Furthermore, machine learning will drastically simplify the management and operation of future 5G campus networks in Industry 4.0 or large agricultural enterprises. This will particularly benefit small and medium-sized enterprises which do not have expert teams.
To introduce machine learning in future networks, there is a great need for the standardisation of architectures, interfaces and data formats, as the standards increase the reliability, interoperability and modularity of a system and its respective components. For this reason, Prof. Stanczak has since 2018 been head of the ITU-T Focus Group "Machine Learning for future networks including 5G", which has just adopted an ITU-T specification on the architecture of machine learning in 5G.
Let's stay with networks: How do you assess Berlin as an ecosystem and centre for technological innovation, especially in the field of artificial intelligence? What distinguishes Berlin from other locations?
Berlin has an outstanding position in this field. There is a very large number of BMBF and BMWi-supported projects as well as orders from industry. This concentration illustrates the high global visibility of Berlin as a location for AI. In addition, we are bringing together the unique German science landscape in Berlin for future projects.
There are already many companies and projects which make use of machine learning and deep learning for their applications. How do you see the willingness to innovate and invest in artificial intelligence in Germany?
For AI to be widely usable in practice and also for safety-relevant or vital applications, further progress in the scientific and technical fundamentals is important. In addition, the data necessary for progress must be available or generated. Here we must implement our ethical ideas in Europe and be at the forefront. Germany has industries and relevant players in all important fields of application.
It is well known that the public discourse on AI fluctuates strongly between the pros and cons of technological progress. How do you see the opportunities, potential and risks for the future?
I am occasionally asked whether AI will cost many jobs. My answer is that I fear that AI will cost us all jobs ... if we do not use the possibilities of AI, but our international competition does. Digital data and their efficient communication and processing, for example with AI, are currently probably the most important scientific and technological innovation mechanisms. By the way, it is precisely these topics - sensor technology, communication, data processing - which the Fraunhofer HHI is dealing with in its research.
What recommendation do you have for people and companies that want to deal with the development of artificial intelligence?
Read, learn and try - as with everything new.
Thanks very much for the conversation.