A Dynamical Microscope for Multivariate Oscillatory Signals: Validating Regime Recovery on Shared Manifolds

Researchers have developed a new AI framework called a 'dynamical microscope' to analyze complex, non-stationary signals like brain activity. It uses an autoencoder to create a latent trajectory model, successfully distinguishing between different dynamical regimes in oscillatory data with high discriminability (η² > 0.5). The method focuses on changes in trajectory speed and geometry on a shared manifold, rather than requiring discrete state separation, providing a new tool for neuroscience and complex systems analysis.