Our benchmark dataset DynaBench is now available from WueData - the research data repository of the Julius-Maximilians-Universität Würzburg.

Earlier this year we reported that our work on DynaBench, A benchmark dataset for learning dynamical systems from low-resolution data, has been published at the machine learning conference ECML-PKDD 2023. Now the data is available for download from our university's research data repository WueData at https://wuedata.uni-wuerzburg.de/radar/de/dataset/sSEeRraAYDgQCgBP

Additionally we also published an easy-to-use python package dynabench which facilitates the utilization of our dataset. The package is available under https://pypi.org/project/dynabench/

Abstract

Previous work on learning physical systems from data has focused on high-resolution grid-structured measurements. However, real-world knowledge of such systems (e.g. weather data) relies on sparsely scattered measuring stations. In this paper, we introduce a novel simulated benchmark dataset, DynaBench, for learning dynamical systems directly from sparsely scattered data without prior knowledge of the equations. The dataset focuses on predicting the evolution of a dynamical system from low-resolution, unstructured measurements. We simulate six different partial differential equations covering a variety of physical systems commonly used in the literature and evaluate several machine learning models, including traditional graph neural networks and point cloud processing models, with the task of predicting the evolution of the system. The proposed benchmark dataset is expected to advance the state of art as an out-of-the-box easy-to-use tool for evaluating models in a setting where only unstructured low-resolution observations are available. The benchmark is available at https://professor-x.de/dynabench.