correlation-modeling

This paper introduces Unicorn, a framework for scalable multi-dataset pretraining on high-dimensional time series that decouples correlation modeling from specific channel identities via a latent prototype codebook, enabling domain transfer and few-shot forecasting.