The authors introduce a new model for multivariate probabilistic time series prediction, designed to address tasks such as forecasting, interpolation, and their combinations. This model is based on copula theory and is an improvement on the recently introduced transformer-based attentional copulas (TACT iS). The new model requires fewer distributional parameters and introduces a training curriculum, resulting in better training dynamics. The authors claim that the new model achieves state-of-the-art performance in diverse real-world forecasting tasks while maintaining the flexibility of the original TACT iS model.
Publication date: 3 Oct 2023
Project Page: Not Provided
Paper: https://arxiv.org/pdf/2310.01327
