Zum Inhalt springen
Technische Universität München

Software

Recent Papers:
chemtrain-deploy: a parallel and scalable framework for machine learning potentials in million-atom MD simulations

P. Fuchs, W. Chen, S. Thaler, J. Zavadlav, Journal of Chemical Theory and Computation 2025, paper , GitHub

chemtrain: Learning deep potential models via automatic differentiation and statistical physics

P. Fuchs, S. Thaler, S. Röcken, J. Zavadlav, Computer Physics Communications, 2025, paper , GitHub

JaxSGMC: Modular stochastic gradient MCMC in JAX

S. Thaler, P. Fuchs, A. Cukarska, J. Zavadlav, SoftwareX 2024, paper , GitHub

An application-agnostic library for stochastic gradient Markov chain Monte Carlo (SG-MCMC) in JAX, enabling Bayesian deep learning.

Published code:
Partial charge prediction trained with AL using Dropout MC

GitHub

Code for partial charge prediction graph neural network trained with active learning using Dropout Monte Carlo.

ML potentials trained on DFT and EXP data

GitHub

Code for training ML potentials concurrently on ab initio and experimental data. 

Relative Entropy Minimization

GitHub

Code for training of coarse-grained molecular models with relative entropy minimization.

Differentiable Trajectory Reweighting

GitHub Zenodo

Code for gradient-based training of molecular dynamics potentials from experimental data.

Jax / Haiku implementation of DimeNet++

GitHub

This DimeNet++ implementation can be used for molecular dynamics simulations with Jax, M.D. and for molecular property prediction.

To top