Tag
This paper introduces a method using composition operators from dynamical systems to linearize and analyze LLM transformations, enabling structural insights and model comparisons through spectral analysis.
DeepMDMD combines deep learning with algebraic constraints to learn compact, dynamically coherent Koopman operator representations that enforce the product rule as an exact constraint. The method outperforms geometric approaches on high-dimensional chaotic and fluid dynamics problems, reducing spectral pollution and enabling stable long-term forecasting.