Enhance VGM Scalability for High-Dimensional Data

Background: Research on time series forecasting models has been actively pursuing improvements in both computational efficiency and forecasting accuracy, particularly when incorporating exogenous variables.

Question / Future Work: Future work should focus on reducing the input dimensionality of the Variate-wise Gating Module (VGM) when dealing with high-dimensional multivariate forecasting tasks where every variable acts as both an endogenous target and an exogenous driver. This is necessary to enhance the model’s scalability and performance in such complex, high-dimensional scenarios.