The paper is at SimTech’s core topic of data-integrated simulation science, combining simulations with data and AI. The authors developed a benchmark for ML for scientific simulations, providing data and the respective generators together with the simulation codes. The set of more than 10 benchmark problems extends the range of PDEs (partial differential equations) in existing benchmarks, adding more realistic and difficult settings, and offers much more extensive data sets than previously accessible. The authors further published their source codes with user-friendly APIs for data and results generation with popular machine learning models. The paper therefore directly targets data reproducibility and makes use of DaRus, the data repository of the University of Stuttgart, to share the benchmark data and simulation codes.
Takamoto, M., Praditia, T., Leiteritz, R., MacKinlay, D., Alesiani, F., Pflüger, D., & Niepert, M. (2022). PDEBench: An extensive benchmark for scientific machine learning. Advances in Neural Information Processing Systems, 35, 1596-1611. (Proceedings of the 36th Conference on Neural Information Processing Systems (NeurIPS 2022)).
Takamoto, Makoto; Praditia, Timothy; Leiteritz, Raphael; MacKinlay, Dan; Alesiani, Francesco; Pflüger, Dirk; Niepert, Mathias, 2022, "PDEBench Datasets", https://doi.org/10.18419/darus-2986, DaRUS, V3
