Tag
CoRe proposes a model-agnostic learning objective for multivariate time-series forecasting that uses frequency coherence and relational graph losses to improve prediction accuracy over standard methods.