MDBench: Benchmarking Data-Driven Methods for Model Discovery

This record contains the dataset accompanying the paper "MDBench: Benchmarking Data-Driven Methods for Model Discovery," accepted to AAAI 2026. MDBench is an open-source benchmark suite for data-driven model discovery, featuring 14 PDE systems and 63 ODE systems generated under multiple noise settings. For full details, see the preprint: https://arxiv.org/abs/2509.20529 and the official codebase: https://github.com/gryaklab/mdbench. The processed archive provides the subsampled noisy and clean datasets in a unified NPZ format. The raw archive contains all original datasets from which the processed versions were derived.