PLAID

PLAID (Physics-informed Learning from AI-aided Descriptions) is an industrial physics-ML benchmark from ICML 2026, targeting real-world industrial simulation scenarios with heterogeneous mesh structures and physical quantities. Its format is CGNS-compatible and plugs directly into mainstream CAE post-processing tools, making it the dataset closest to production for accelerating industrial simulation (surrogate / reduced-order modeling) with ML.

Use cases: Validating surrogate-model accuracy on complex industrial geometries; connecting with mainstream CAE toolchains for reduced-order-model verification.