Stellenangebot / Job offer
2 Postdoc Positions in Deep Learning for Earth System Science
The Model-driven Machine Learning group within the Institute of Coastal Systems – Analysis and Modeling at Helmholtz-Zentrum Hereon, seeks two curious and driven postdocs to start as soon as possible. They will work at the absolute forefront of AI for science, with scalable modern methods, interdisciplinary expertise and industrial-scale computation. A solid background in deep learning is required, including some published work. Expertise in Earth system science and numerical simulation is welcome but optional.
Cooperating with Forschungszentrum Jülich, Alfred-Wegener-Institut and Karlsruhe Institute of Technology, we will design, train, test and deploy HClimRep, the first truly climate-capable Earth system foundation model, on observed and simulated climate data. This transformer-based model will be used for simulation, forecasting, superresolution, bias correction and counterfactual exploration. One postdoc will focus on long-term accuracy and the other on stratospheric advection, but both will have considerable freedom to innovate and will participate in designing, training, testing and deploying HClimRep.
The Model-driven Machine Learning group combines ML- and physics-based approaches to Earth system modeling. Our work schedule combines remote and in-person interactions. The successful candidates will focus on challenging and impactful research, with minimal administrative or teaching duties. The positions are initially limited to 3 years.
Your tasks:
- develop and train HClimRep on climate data ranging from the oceans to the stratosphere
- evaluate and tune specialized techniques for long-term accuracy and stability (position 1)
- train a self-attention-based module for tracer advection (position 2)
- write, publish and present research in scientific journals and conferences
Your profile:
- PhD in mathematics, physics, machine learning, computer science or the geosciences
- can carry out research independently with curiosity, creativity, initiative and persistence
- works well in a small group with diverse skill sets and backgrounds
- learns quickly on the job and from research articles on ML and the geosciences
- solid publication record showing creative ideas and hard work. High-profile journal articles are welcome but not required
- comfortable programming in Python and PyTorch
- can communicate well in spoken and written English
