Meet the Helmholtz AI local unit for matter @ HZDR
At Helmholtz AI we aim to push boundaries in terms of artificial intelligence and machine learning, but this still requires some human-powered efforts. Let us introduce the ‘humans’ who make Helmholtz AI possible!
This month, we spoke with the seven humans who currently compose the Helmholtz AI local unit for research field matter at Helmholtz-Zentrum Dresden-Rossendorf (HZDR).
Nico Hoffmann, young investigator group leader. Nico earned his PhD in 2016 from Technische Universität Dresden in medical image analysis. He developed statistical machine learning methods that compensate motion artefacts and advanced semiparametric regression models as well as neural networks for intraoperative recognition of evoked neuronal activity of the exposed human brain. He visited the Laboratory of Mathematics in Imaging of Harvard University from 2018 to 2019. During that time, he developed recurrent convolutional neural networks for reconstruction of nerve fibre bundles of the human brain. Right before Helmholtz AI, Nico was with the Computational Radiation Physics group of Helmholtz-Zentrum Dresden-Rossendorf.
His Helmholtz AI research group is advancing physics-informed surrogate models for very fast simulation as well as online comprehension of future particle accelerators. Extending these surrogate models by a reliable analysis of experimental diagnostics using Model-aware neural networks finally paves the way for future data-driven but still explainable comprehension of fundamental physics. Applications of our methods basically range from resolving ultra fast processes of Laser-Ion acceleration, extreme conditions in Astrophysics and very small structures of biomolecules.
Peter Steinbach, leader of the Helmholtz AI consultant team at HZDR. Peter’s professional life took some dents here and there. After school, he was interested in the origins of the universe (‘nothing more humble than that!’) and so he went into physics. He pursued his Diploma at DESY in experimental particle physics and subsequently moved to do a PhD in the same field, but at TU Dresden. During this time working on the first LHC data, he came into contact with tons of statistics and high performance computing (HPC). After graduating with a PhD on a ‘far too complicated topic for this paragraph’, he went into a semi-industry, semi-academia job being a research software engineer with a focus on HPC. ‘As this ‘Deep Learning’ became a thing in science some years ago, I dived into that subject and was happy to return to Data Science with it. So here I am... doing Machine Learning and Data Science again - lucky me!’
As leader of the consultants team in matter, he splits his day job in parts on administrative tasks, doing his best to lead the team and assist in strategic decision making, while taking care of vouchers and reading many papers on new methods. Put plainly, Peter spends his time adapting new or existing Machine Learning methods to solve pressing scientific problems of colleagues and community members.
Steve Schmerler, a Helmholtz AI consultant at HZDR. Steve holds a PhD in theoretical physics, where he did a lot of HPC and scientific computing. Then he spent some time at a cloud computing company as a software engineer before joining Helmholtz AI. As a consultant of the HZDR unit, he is currently working on PyTorch-enabled multivariate polynomials in the context of Physics-Informed Neural Networks, in collaboration with the Young investigator group of the unit.
Neda Sultova, a Helmholtz AI consultant at HZDR. Neda studied computer science, but she is also passionate about drawing.At Helmholtz AI, she puts her skills in machine learning engineering, data preprocessing and tooling into practise.
Sebastian Starke, a Helmholtz AI consultant at HZDR. Sebastian studied mathematics and statistics at the OvGU Magdeburg before going back to Dresden to work as an algorithm engineer for speech recognition tasks. After that, he joined the HZDR and started to pursue his PhD, where he applies deep learning techniques to medical images to build statistical risk models for cancer patients. Now, he supports researchers who work in the field of matter by providing advice on suitable ML techniques for their data and research questions and help in implementing them.
Patrick Stiller, a PhD student at Nico Hoffmann’s group. In 2013, Patrick started his studies in computer science. Since his first semester, he was interested in artificial intelligence, so he chose artificial intelligence, computer vision and computational biology as major subjects. In his sixth semester he started his job at HZDR, where he got his first hands on experience on machine learning and neural networks. He has also been involved in the machine learning community Dresden, an organization for the democratization of machine learning knowledge. As part of Helmholtz AI, Patrick will focus on Physics-informed surrogate models for PDE-driven simulations. He wants to build and study fast, generic and robust surrogate models which enables fast parameter scans.
Yichao Liu, a research assistant at Nico Hoffmann’s group. Yichao spent four months in DFKI with the Autonomous driving group working on a Neural Architecture Search (NAS) study. He extremely reduced the existing model size of a NAS method-DARTS by diversifying the layer structure without precision loss. Before that, during his master stage, he developed a graph-based autoencoder to improve the expressiveness of classical autoencoders. He also has some time series data processing experience on lane change prediction of autonomous driving cars. At Helmholtz AI, he is responsible for SAXS data reconstruction. His main task is to speed up GISAXS data analysis by different kinds of autoencoders.
Creative solutions across disciplines
Neda thinks Helmholtz AI provides ‘room for creative solutions and approaches, and plenty of opportunity to learn’, an opinion Sebastian and Yichao share. Steve adds: ‘interesting problems, time to develop solid solutions, good computational resources’ to the mix.
For Nico and Patrick, it is all about interdisciplinarity. Patrick really likes ‘to be a part of this network and learn from the expertise of the other members’. ‘This collaborative effort basically combines the strengths of all members of the Helmholtz Association’, adds Nico.
Peter is ‘super grateful that I get paid for extracting knowledge out of data using ML, statistics or simply clever guessing - this is what I love doing. Moreover, I am in the fortunate situation working with a wonderful, diverse and intelligent team in a department that breaks ground with a modern and agility-focussed approach to projects.’
Cultural change to tackle great challenges
In the future, Nico believes ‘we will be sharing our experiences with data-driven methods -as well as code!- and might potentially approach generic methods applicable for tackling the grand challenges of our world’. Sebastian agrees: ‘I hope that the network will grow and flourish and at the same time, we, as a community will grow closer, allowing for fruitful exchange and support beyond the borders of the individual centers.’
As lead of the consultant team, Peter hopes that ‘we can support the cultural change to a point where science is boosted by Helmholtz AI's transformative service-oriented commitment. Something that has worked wonders in the life sciences and other fields - why not in the whole Helmholtz association?’
Professional life always has some impact in personal endeavours. Neda, for example, admits that ‘people would be surprised to know that I can talk for hours on every of the above topics, if asked. Steve’s personal philosophy is that ‘everything is an optimization problem’, while Patrick’s is to ‘never stop learning, because life never stops teaching’.
Peter is inspired by Ludwig van Beethoven for ‘being able to put his inner passions (both with a happy and sad colour) in the high art of composing classical music to express himself. Beethoven created complex art that remained approachable and inspiring over centuries. Breathtaking every time I hear it!’. His personal philosophy is to ‘never take myself and others too seriously! Life is too short not to laugh’.
Nico enjoys outdoor climbing and bouldering: ‘you basically become one with nature when thinking about how to climb "that rock". The view and feeling once you make it is truly amazing!’ Sebastian’s is also into the outdoors; in his free time he likes to go bike riding. On the other hand, Yichao likes sitting in front of Brühlsche Terrasse and enjoying the wind blowing.