An-Nan Yu

Generative time series forecasting models can now produce complete future trajectories with competitive marginal accuracy and superior path-level performance, outperforming quantile-based methods on trajectory-dependent metrics.