<ul><li>CrossLinear is a new Linear-based forecasting model designed for time series forecasting with exogenous variables.</li><li>It addresses challenges in modeling dependencies between variables by incorporating a plug-and-play cross-correlation embedding module.</li><li>The module captures dependencies with minimal computational cost, differentiating between time-invariant/direct and time-varying/indirect dependencies to prevent overfitting.</li><li>Experiments show that CrossLinear outperforms traditional models in short-term and long-term forecasting tasks, with the cross-correlation embedding module proving effective.</li></ul>

CrossLinear: Plug-and-Play Cross-Correlation Embedding for Time Series Forecasting with Exogenous Variables

Discover more