Esther Heid

Do you love building AI/ML models across a wide range of research topics? TU Wien is hiring two senior scientists/postdocs for a university-wide AI/ML service center (www.tuwien.at/datatudiscovery).

Computer science: jobs.tuwien.ac.at/Job/263874
Science/engineering: jobs.tuwien.ac.at/Job/263875

Feynman-Kac-Flow: Inference Steering of Conditional Flow Matching to an Energy-Tilted Posterior Conditional Flow Matching(CFM) represents a fast and high-quality approach to generative modelling, but in many applications it is of interest to steer the generated samples towards precise requiremen...

Want to steer your flow matching models, not just diffusion? We have just derived Feynman-Kac-Steering for flow matching, eg to steer chemical transition states toward correct chirality. Fantastic work by Konstantin Mark, along with Leonard Galustian and Maximilian Kovar. arxiv.org/abs/2509.01543

Graph-based prediction of reaction barrier heights with on-the-fly prediction of transition states The accurate prediction of reaction barrier heights is crucial for understanding chemical reactivity and guiding reaction design. Recent advances in machine learning (ML) models, particularly graph ne...

Graph neural networks are inherently limited in predicting chemical reaction properties by the absence of 3D information. We showcase how generative models can create guess structures for reaction pathways, and encode these into GNNs for reaction barrier height prediction: doi.org/10.26434/che...

Definitely, it would make a great stress test. Feel free to come by anytime.

2) Yet, previous models such as diffusion models are slow to train and use, and not very accurate moreover. Here, flow matching comes into play allowing for a 100-fold speed-up. In combination with recent developments in machine learning potentials, GoFlow furthermore achieves higher accuracies.

1) Given a chemical reaction, can we predict how exactly the atoms move to go from reactants to products? This reaction pathway determines the barrier height and rate of the reaction, but it is utterly expensive to compute using quantum mechanics. Here, generative AI can help.

GoFlow: Efficient Transition State Geometry Prediction with Flow Matching and E(3)-Equivariant Neural Networks Transition state (TS) geometries of chemical reactions are key to understanding reaction mechanisms and estimating kinetic properties. Inferring these directly from 2D reaction graphs offers chemists ...

Very proud to introduce the first preprint of the Heid lab on generative AI for chemical reactions, where we combine flow-matching with equivariant neural networks to predict transition state geometries only based on reaction graphs. Read more below or check out the paper:

doi.org/10.26434/che...