<ul><li>A new causal discovery algorithm called Kausal has been developed using a deep Koopman operator-theoretic formalism.</li><li>Standard statistical frameworks like Granger causality face challenges in quantifying causal relationships in nonlinear dynamics due to complex feedback mechanisms and nonstationarity.</li><li>Kausal leverages Koopman operator methods to approximate nonlinear dynamics in a linear space of observables and uses deep learning to infer optimal observables for causal analysis.</li><li>Numerical experiments show that Kausal outperforms existing approaches in discovering and characterizing causal signals, with application potential in real-world phenomena like El Niño-Southern Oscillation.</li></ul>

Deep Koopman operator framework for causal discovery in nonlinear dynamical systems

Discover more