iCreatAI News
Skip to content
MIT News AI· David Chandler | Laboratory for Information and Decision Systems·· 2 hr agoAI score45

New approach enables physically consistent AI models of chaotic systems

New approach enables physically consistent AI models of chaotic systems

AI summary

MIT LIDS researchers developed a method that adds energy-based hard constraints and a projection layer to machine learning models of chaotic systems, providing theoretical guarantees that predictions stay bounded over arbitrarily long forecast windows. Tested on systems from simple textbook cases to complex multidimensional ones, the approach matched true outcomes while conventional methods drifted into impossible predictions. The work, published in PNAS with open-source code, is model-agnostic and the team plans to test it on larger systems including realistic weather models.

Source: MIT News AI · news.mit.edu