Skip to content

Modeling Local Cross-Variate Correlations

Home / Open Questions / Modeling Local Cross-Variate Correlations

Background: Time series forecasting models often struggle to effectively capture inter-variate correlations at a local, fine-grained temporal resolution.

Question / Future Work: Future research is needed to develop methods that explicitly incorporate local-level cross-variate correlations within hybrid architectures, as current approaches often rely on global context to capture such interactions, potentially missing relevant local dependencies.

Why It Matters: This represents a significant architectural bottleneck in multivariate time series forecasting that prevents models from capturing the complex, localized interaction patterns between different variables.

Metadata & Links