André Graubner

Tsinghua University / ETH Zurich

prof_pic.jpg

I am interested in building genetic and epigenetic foundation models, for example with State Space Models.

Before that I built tools, methods and datasets to advance research questions in climate science at Nvidia and the European Space Agency. When appropriate, I like to apply and improve methods from deep learning.

During my undergrad at ETH Zurich, I worked with my supervisor Nora Hollenstein to bridge rationale-based interpretability methods for natual language processing and psycholinguistic data like eye-tracking.

I also spent time as a part-time researcher at Lawrence Berkeley Lab working on large-scale deep learning for climte analytics, supervised by Prabhat and Karthik Kashinath. As a follow-up, I co-led the creation of the largest hand-crafted extreme weather dataset for the European Space Agency and built deep learning models on top of it.

Afterwards, I joined Nvidia as a research intern under supervision of Anima Anandkumar, improving calibration of the FourCastNet neural weather model.

Currently I am a Masters student at Tsinghua University in Wenwu Zhu’s lab, connecting abstraction-based library learning, program synthesis and mathematical reasoning to enable machines to re-organize formal theories based on experience, allowing for better generalisation and interpretability.

news

No news so far...

selected publications

  1. calibration.png
    Calibration of Large Neural Weather Models
    Andre Graubner, Kamyar Azizzadenesheli, Jaideep Pathak, Morteza Mardani, Mike Pritchard, Karthik Kashinath, and Anima Anandkumar
    In NeurIPS Workshop on Tackling Climate Change with Machine Learning, 2022
  2. climatenet.png
    ClimateNet: an expert-labeled open dataset and deep learning architecture for enabling high-precision analyses of extreme weather
    Prabhat, K. Kashinath, M. Mudigonda, S. Kim, L. Kapp-Schwoerer, A. Graubner, E. Karaismailoglu, L. Kleist, and 15 more authors
    Geoscientific Model Development, 2021
  3. segmentation.png
    Spatio-temporal segmentation and tracking of weather patterns with light-weight Neural Networks
    Andre Graubner, Lukas Kapp-Schwoerer, Sol Kim, and Karthik Kashinath
    In NeurIPS Workshop on AI for Earth Sciences, 2020