AdaSpecK: Spectral-Koopman Framework for Temporal Domain Generalization
October 5, 2026
AdaSpecK uses spectral-regularized Koopman dynamics and target-conditioned attention to model non-stationary streaming data. It extracts denoised low-frequency trajectories in a linearized latent space to improve generalization across irregularly sampled domains.
HOW THIS AFFECTS YOU
●
researcherYou can use this method to handle distribution shifts in time-series data without overfitting to domain-specific noise.