N-HiTS: Neural Hierarchical Interpolation for Time Series Forecasting

The NHITS model enhances the multi-step forecasting strategy by incorporating innovative hierarchical interpolation and multi-rate data sampling techniques inspired by wavelet analysis.

It assembles its predictions sequentially and emphasizes its components with different frequencies and scales. NHITS significantly improves accuracy in long-horizon forecasting tasks while reducing computation time by orders of magnitude compared to existing neural forecasting approaches.