In its second funding round for Helmholtz AI projects, Helmholtz is investing 6.6 million euros in collaborative research projects in the field of applied artificial intelligence and machine learning.
Helmholtz AI strengthens the application and development of applied artificial intelligence (AI) and machine learning (ML). In the second call for projects of Helmholtz AI, a top-class international panel of experts selected 17 collaborative high-risk, high-reward research projects. Helmholtz is investing a total of 6.6 million euros in these novel approaches – half of which is provided by the Association’s Initiative and Networking Fund.
“This year marks the bicentennial birthday of Hermann von Helmholtz, and this is a great opportunity to apply artificial intelligence and machine learning strategies to some of today’s grand challenges”, says Otmar D. Wiestler, the President of the Helmholtz Association. “The 17 selected projects use powerful AI and machine learning tools to address complex issues in their respective areas. I wish our scientists success and outstanding research results.”
62 project proposals were submitted in the second Helmholtz AI project call (up from 55 in the first round), demonstrating that this particular funding line is in persistent high demand. The 17 selected projects use novel analytical tools to solve pressing scientific and specific transfer challenges using AI. The projects promote the testing of these new approaches, are supported by several partners and will be implemented in up to three years – with the potential to quickly spawn larger follow-up projects.
“The Helmholtz AI projects are at the core of our mission: they build bridges between Helmholtz Centers and open up opportunities to apply AI to the many unique datasets across all research fields,” says Fabian Theis, the scientific director of Helmholtz AI. “Pursuing these joint approaches weaves a network between disciplines and I am delighted to see our open and dynamic community growing further.”
With the Helmholtz AI platform, Helmholtz is investing up to 11 million euros annually in six innovative research units and a transdisciplinary network for applied AI. This network builds on the community's strengths in AI and its unique data sets. In this way, Helmholtz AI contributes to solving the grand challenges of our time, fostering cross-field and cross-center collaboration, also with external university- and industry partners. The network keeps growing – the next call for proposals will be open soon: from 1 September until 1 December 2021.
Selected AI Projects 2020
The following 17 projects will receive funding of up to €400,000 each over a total funding period of two to three years.
- SynRap: Machine-learning based synthetic data generation for rapid physics modeling (DESY + HZDR)
SynRap investigates the generation of simulated (“synthetic”) data using surrogate models, which will be used in a second step for efficient training of neural networks. A unified surrogate model framework will be developed and used to tackle common challenges in two different research areas – high-energy physics (HEP) and high energy-density (HED) phenomena.
Contact: Isabell-Alissandra Melzer-Pellmann, firstname.lastname@example.org (DESY)
- PSDAI: Patient-specific diagnostic AI-systems via one-shot domain adaptation (DKFZ + DLR)
Patient-specific diagnostic AI-systems via one-shot domain adaptation deep learning has shown its potential in medical image analysis, but the lack of adequate generalization impedes the translational step to the clinic. This project aims to develop an approach that uses patient-specific information at inference time to improve generalization for that particular setting.
Contact: Titus Brinker, email@example.com (DKFZ)
- GANCSTR: GAN-based calculation of concentrated radiation on solar tower receivers (DLR + FZJ)
Within GANCSTR machine learning methods are developed for very accurate prediction forecasts in solar thermal power plants. Such power plants consist of a large number of mirrors, so-called heliostats, which must be optimally controlled in order to achieve a very low levelized cost of energy. Generative adversarial networks are used to determine the distribution of solar radiation reflected from each heliostats and maximize the power plant's energy output.
Contact: Daniel Maldonado Quinto, firstname.lastname@example.org (DLR)
- FoAIm: Coupling of complex multiscale structural foam characteristics with AI based methods (DLR + KIT)
The long term vision of this project is to accelerate the characterization of material properties and thus considerably facilitate the development of new materials and components e.g. for future vehicles. As a first step the project partners will develop a ML architecture including explainable AI approaches to fulfill the tasks of characterizing mechanical properties from microstructure images and tailoring microstructures to desired macroscopic behavior for structural foams.
Contact: Alexander Greß, email@example.com (DLR)
- DeGen: Deep Generative Models for Causal Prognosis using Neuroimaging Data (DZNE + FZJ)
Contact: Martin Reuter, firstname.lastname@example.org (DZNE)
- T^6: ReacTive TransporT experimenT digiTal Twin (FZJ + GFZ)
Reactive Transport Digital Twin (T6) develops an expanded toolbox for experimentalists integrating computer vision and AI/ML methods for ultra-fast forward coupled hydro-geochemical simulations, enabling real-time assessment and correction of microfluidic experiments.
Contact: Jenna Poonoosamy, email@example.com (FZJ)
- SuperPI: Deep-learning empowered super-resolution plankton imaging (GEOMAR + HMGU)
The goal of SuperPI is to provide AI-based solutions for improving the optical resolution of underwater plankton imaging. Particularly, developing algorithms for novel "enhanced depth-of-field" imaging technology will allow to combine high optical resolution in large sample volumes. Thus, the project is envisioned to push the boundaries of plankton imaging applications, as well as optical microscopy in general.
Contact: Jan Taucher, firstname.lastname@example.org (GEOMAR)
- PCDL-QuaSPA: Physics-constrained deep learning framework for quantifying surface processes across the Arctic region (GFZ + AWI)
The permafrost-laden landscape of the Arctic is highly susceptible to degradations in the warming climate, and harbours the potential to exacerbate climate change due to its huge store of soil organic carbon. Large-scale monitoring and fast predictive simulations of permafrost-related features and natural systems are thus urgent and important. The project aims to develop both a deep-learning model capable of detecting and quantifying permafrost-landscape changes, and a physics-informed deep-learning framework to enable rapid modelling of complex Arctic surface-processes systems.
Contact: Hui Tang, email@example.com (GFZ)
- AI4GNSSR: Artificial Intelligence for GNSS Reflectometry: Novel Remote Sensing of Ocean and Atmosphere (GFZ + DLR)
Contact: Milad Asgarimehr, firstname.lastname@example.org (GFZ)
- DEEPROAD: AI-based development of next-generation diagnostics and precision medicine for Alzheimer’s disease (HMGU + DZNE)
Contact: Ali Ertürk, email@example.com (HMGU)
- AI²: Eyesight to AI: Discovery of efficient corrosion modulators via predictive machine learning models (Hereon + HMGU)
The fundamental concept of the project is to develop a pattern recognition routine that enables high-throughput quantification of the effect of small organic additives on the degradation of a magnesium alloy by automated classification of corrosion imprints. Subsequently, the quantified optical signal will be used as a target parameter for different supervised and semi-supervised learning approaches to predict the performance of untested additives. Accuracy and robustness of the developed models will be validated by experimental blind tests.
Contact: Christian Feiler, firstname.lastname@example.org (HZG)
- AMR-XAI: Crushing antimicrobial resistance using explainable AI (HZI + CISPA)
Antimicrobial resistance is perhaps among the most urgent threats to human health. AMR-XAI proposes to learn a small set of easily interpretable models that together explain the resistance mechanisms in the data using statistically robust methods for discovering significant subgroups. Key to our success will be the tight integration of domain expertise into the development of the new algorithms and early evaluation on real-world data.
Contact: Olga Kalinina, email@example.com (HZI)
- AI-InSu-Pero: Artificial intelligence guided in-situ analysis of scalable perovskite thin film deposition (KIT + DKFZ + FZJ)
As much as AI methods are already advancing some pioneering fields such as medical image diagnostics, they are suited to revolutionize combinatorial materials science and processing. In this project, AI algorithms are developed that are required for the detection of defects and inhomogeneities as well as film quality correlations in in-situ image data of solution-processed perovskite thin films.
Contact: Ulrich Wilhelm Paetzold, firstname.lastname@example.org (KIT)
- GLAM: Generative lung architecture modeling (MDC + HMGU)
Complex 3D tissue disease models are being used as surrogates for pre-clinical drug development. However, current methods for fabricating artificial tissues are based on image-derived, or manually designed tissue architectures, and lack high-throughput applicability. This project is developing generative methods for designing bio-printable lung tissues across a spectrum of disease severity in the specific context of mouse and human lung disease.
Contact: Kyle Harrington, Kyle.Harrington@mdc-berlin.de (MDC)
- CausalFlood: Identifying causal flood drivers (UFZ + DLR)
Contact: Jakob Zscheischler, email@example.com (UFZ)
- PAI: Pollination artificial intelligence (UFZ + DLR)
Pollination artificial intelligence (PAI) will transfer existing large-scale applicable AI-methods to the field of pollination ecology. PAI will train a hierarchical deep learning approach capable of identifying thousands of European pollinators from in-field observations, by using existing image databases and expert knowledge attributes. The successfully trained AI model will have its performance systematically tested in field studies.
Contact: Tiffany Knight, firstname.lastname@example.org (UFZ)
- XAI-graph: Explainable AI and improved measurements of uncertainty for machine learning on (biomolecular) structure graphs (UFZ + HZI)
The goal of XAI-graph is to increase the credibility of predictive approaches in toxicology by introducing explainability into existing AI approaches and by developing methods to quantify uncertainty of AI based predictions. The aim is to i) promote these principles in toxicology and transfer them into applications to support the prioritization of chemicals for regulatory testing, and ii) to gain new functional/biological insights through the investigation of decisive features.
Contact: Jana Schor, email@example.com (UFZ)