Networking data and knowledge
Medicine, climate research, aerospace: the variety of disciplines for which AI plays a role is also reflected within the Helmholtz Association. In order to support cooperation and the exchange of knowledge and data within the research network, the association is currently setting up an interdisciplinary platform for which it provides long-term funding of € 11.4 million annually – the Helmholtz Artificial Intelligence Cooperation Unit (HAICU).
“As the central unit, Helmholtz Zentrum München links several Helmholtz centres that form local units with thematic focuses. Jülich covers the area of key technologies/information in particular,” says Dr. Timo Dickscheid, AI expert at the Institute of Neuroscience and Medicine (INM-1). “Each local unit consists of a fully equipped young investigator group and a high-level support team that supports other scientists in their projects with AI expertise.”
Dickscheid’s research group “Big Data Analytics” will play an important role for the Jülich segment. It creates a highly accurate model of the human brain, the resolution of which will reach down to individual nerve cells. To this end, the group is further expanding deep learning methods in order to analyse microscopic image data at a magnitude of several terabytes. AI is to learn to automatically detect microstructures, map brain areas and assemble thousands of tissue sections into 3-D views of the brain. The group wants to develop solutions that can also be used in other research areas – such as learning with only a few training examples.
The research group “Cross-Sectional Team Deep Learning” of the Jülich Supercomputing Centre (JSC), headed by Prof. Morris Riedel and Dr. Jenia Jitsev, is also involved in the Jülich unit of HAICU. Together with the new high-level support team of the JSC, it will promote and support research into machine learning and deep learning. The focus will be on large-scale, adaptive neural networks. They can learn from a continuous stream of data over weeks or months and transfer what they have learned to various tasks that consist of only a few examples. With the help of this transfer learning, complex classical simulations in physics, for example, can be improved. For this purpose, the learning neural networks are coupled to the simulations in order to further optimise their models by means of generated and observed data.
Photo: Forschungszentrum Jülich/Ralf-Uwe Limbach