QiYao-I: A Manifold Based Foundation Model for Irregular Multivariate Time Series Forecasting

Irregular multivariate time series forecasting is a challenging yet important problem in real-world applications, where observations are often irregularly sampled and asynchronously recorded across variables.