Spatio-temporal fourier transformer for long term dynamics prediction

Area: Artificial intelligence

CASS member: RAPIDS

Description

StFT (Spatio-temporal Fourier transformer) is a novel machine learning model developed to emulate long-term dynamics of multi-scale and multi-physics systems. The model overcomes the limitations of rapid error accumulation, particularly in long-term forecasting of systems characterized by complex and coupled dynamics. StFT achieves outstanding accuracy and computational efficiency by effectively capturing multi-scale interactions, and quantifies the uncertainties inherent in the predictions. The model StFT leverages a structured hierarchy of StFT blocks, and explicitly captures dynamics across both macro- and micro- spatial scales. Evaluations conducted on three benchmark datasets (plasma, fluid, and atmospheric dynamics) demonstrate the advantages of our approach over state-of-the-art ML methods.

Target audience

Machine Learning researchers who are interested in neural operators, and domain scientists interested in spatio-temporal forecast and multi-scale modeling.

License: BSD-3-Clause

Additional resources