The slides are available on Google Drive under this link.
This repository contains the code for the tutorial on Atomistic Adversarial Attacks and QUESTS: Quick Uncertainty and Entropy from STructural Similarity presented to the UChicago AI + Science Summer School 2024. The notebooks are available in Google Colab at the following links:
- Atomistic Adversarial Attacks: https://bit.ly/UChicago-AtomAttacks
- QUESTS: https://bit.ly/UChicago-Entropy
The tutorials have some precomputed information in folder named results and datasets in the folders gap20 and rmd17. Only part of the datasets are available in this repository for brevity.
The references for the papers are:
-
D. Schwalbe-Koda*, A. R. Tan*, and R. Gomez-Bombarelli. Differentiable sampling of molecular geometries with uncertainty-based adversarial attacks. Nat. Commun. 12, 5104 (2021). Link
-
D. Schwalbe-Koda et al. Information theory unifies atomistic machine learning, uncertainty quantification, and materials thermodynamics. arXiv:2404.12367 (2024). Link
The full references for the datasets are:
- GAP-20: P. Rowe, V. L. Deringer, P. Gasparotto, G. Csányi, and A. Michaelides. "An accurate and transferable machine learning potential for carbon." The Journal of Chemical Physics 153 (2020). DOI: https://doi.org/10.1063/5.0005084. Dataset: https://www.repository.cam.ac.uk/items/ac362ef1-fa35-4bd4-8b56-2fde6d7b1d2c
- MD-17: S. Chmiela, A. Tkatchenko, H. E. Sauceda, I. Poltavsky, K. T. Schütt, and K.-R. Müller "Machine learning of accurate energy-conserving molecular force fields." Science Advances 3 (2017). DOI: https://doi.org/10.1126/sciadv.1603015.
- rMD-17: A. Christensen and O. A. von Lilienfeld. "On the role of gradients for machine learning of molecular energies and forces." Machine Learning: Science and Technology DOI: https://doi.org/10.1088/2632-2153/abba6f Dataset: https://figshare.com/articles/dataset/Revised_MD17_dataset_rMD17_/12672038
This repository is made available under the BSD-3 License, with the exception of the parts of the datasets:
- The rMD-17 dataset is distributed under the Public Domain.
- The GAP-20 dataset is distributed under the CC-BY 4.0 License.