Deep Learning

Vision-LSTM -- xLSTM as Generic Vision Backbone

We introduce Vision-LSTM (ViL), an adaption of the xLSTM building blocks to computer vision.

Aurora -- A foundation Model of the Atmosphere

Aurora leverages the strengths of the foundation modelling approach to produce operational forecasts for a wide variety of atmospheric prediction problems, including those with limited training data, heterogeneous variables, and extreme events

xLSTM -- Extended Long Short-Term Memory

How far do we get in language modeling when scaling LSTMs to billions of parameters, leveraging the latest techniques from modern LLMs, but mitigating known limitations of LSTMs?

Geometry-Informed Neural Networks

We introduce geometry-informed neural networks (GINNs) to train shape generative models without any data.

Universal Physics Transformers -- A Framework For Efficiently Scaling Neural Operators

We introduce Universal Physics Transformers (UPTs), an efficient and unified learning paradigm for a wide range of spatio-temporal problems. UPTs operate without grid- or particle-based latent structures, enabling flexibility and scalability across meshes and particles.

Mim-refiner -- A contrastive learning boost from intermediate pre-trained representations

We introduce MIM (Masked Image Modeling)-Refiner, a contrastive learning boost for pre-trained MIM models.

We identify particle clustering originating from tensile instabilities as one of the primary pitfalls. Based on these insights, we enhance both training and rollout inference of GNN-based simulators with varying components from standard SPH solvers, including pressure, viscous, and external force components.

Smoothed particle hydrodynamics (SPH) is omnipresent in modern engineering and scientific disciplines. SPH is a class of Lagrangian schemes that discretize fluid dynamics via finite material points that are tracked through the evolving velocity …

Lie Point Symmetries and Physics-Informed Networks

We present how to use Lie Point Symmetries of PDEs to improve physics-informed neural networks. Published at NeurIPS 2023.

Data-Driven Simulations

I am firmly convinced that AI is on the cusp of disrupting simulations at industry-scale. Therefore, I have started a new group at JKU Linz which has strong computer vision, simulation, and engineering components. My vision is shaped by experience both from university and from industry.

PDE-Refiner - Achieving Accurate Long Rollouts with Neural PDE Solvers

PDE-Refiner is an iterative refinement process that enables neural operator training for accurate and stable predictions over long time horizons. Published at NeurIPS 2023 (Spotlight).