Forschungszentrum Jülich is Developing New AI Foundations Models to Advance Scientific Applications
There are huge volumes of data being generated in the field of science. However, the full potential of the data can only be realized if scientists can analyze ever-increasing amounts of data. The recent advances in the field of AI offer an opportunity to tackle this challenge.
Forschungszentrum Jülich, together with its partner institutions within the Helmholtz Association, is on a mission to build a new generation of AI foundation models for science. Several pilot projects have been launched to take the application of AI in science to a new level.
The foundation models are AI neural networks trained on large-scale datasets to handle a wide variety of tasks. They serve as the backbone or core architecture for various natural language processing (NLP) tasks. Foundation models are significantly more adaptable than traditional AI models, making them suitable for scientific applications.
Forschungszentrum Jülich is a German national research institution that conducts research on a variety of areas including health, energy, environment, information technology, and more. The center often collaborates with other research institutions, universities, and industry partners to address some of the most pressing challenges through scientific innovations.
The new projects are part of the newly established Helmholtz Foundation Model Initiative (HFMI), which is set to receive around $23 million euros from the Helmholtz Association over a period of three years.
“We are convinced that with foundation models, we can push the boundaries of science. Helmholtz not only brings outstanding talents and comprehensive datasets from various research areas but also brings together a unique computer infrastructure,” says Otmar Wiestler, President of the Helmholtz Association.
One of the foundation models, HClimRep, is geared toward climate research. It will be a deep-learning model, capable of conducting complex simulations using data from the ocean, atmosphere, and sea ice to allow us to better understand our planet’s climate and develop solutions to address the devasting impact of climate change.
For us to respond to climate change, we need a deeper understanding of the global carbon budget. Scientists have long struggled to analyze scattered data and quantify how changes in vegetation and soil impact CO2 sources and sinks.
Using data from various sources such as satellites, drones, or local CO2 monitoring stations, the 3D-ABC foundation model will allow scientists to capture key parameters of the global carbon cycle of vegetation and soil.
AI will also be leveraged to analyze new data and findings in material science, enabling researchers to accelerate the implementation of innovative solar cell concepts. This project, named SOL-AI, is aimed at the development and optimization of photovoltaic materials.
Photovoltaics is a renewable energy technology that converts sunlight into electricity using semiconductor materials. The SOL-AI foundation model is expected to develop solutions that will have practical relevance for both scientific research and industry.
The fourth project is called the Synergy Unit. It is designed to develop, deploy, and connect foundation models. While other projects focus on specific issues, the Synergy Unit addresses overarching questions about all participating projects such as the scalability or training of datasets.
The Synergy Unit aims to not only provide clear value for science but to also ensure the long-term impact of the Helmholtz Foundation Model Initiative by making the final results of foundation models available to the public as open source. This includes the training data, code, and training model. These projects demonstrate the immense potential of AI to revolutionize the field of science, transcending the limitations of traditional methods and breaking barriers to greater innovation.
Related Items
NVIDIA CEO Hails New Data Science Facility As ‘Starship of The Mind’
The Rise of the Industrial Data Scientist in an Industrial AI World
Pecan AI Leaps Over the Skills Gap to Enable Data Science On Demand
