MIT team keeps AI models of chaos within physical limits
Ongoing: AI Hot

Researchers at MIT have found a way to stop machine learning models of chaotic systems, such as weather and turbulent flows, from drifting into physically impossible predictions. The peer-reviewed paper appears in the Proceedings of the National Academy of Sciences (PNAS).
Graduate student Sunbochen Tang and professors Navid Azizan and Themistoklis Sapsis built the method. The model learns an energy-like function and a boundary for its chaotic behaviour, and a added layer forces energy to fall outside that boundary. The team says this gives formal guarantees that predictions stay bounded, and it is making the information to use the approach freely available. The findings come from the university's own news office, and the results are not independently assessed here.