Scaleformer: Iterative Multi-scale Refining Transformers for Time Series Forecasting

[StatMixedML]

Since you already have most of the base-models available, iteratively refining forecasts at multiple scales as suggested by Scaleformer: Iterative Multi-scale Refining Transformers for Time Series Forecasting can potentially further improve accuracy.

You can find the repo here https://github.com/Scaleformer/Scaleformer.

I am not exactly sure how this relates to temporal forecast reconciliations, but having additional information from several temporal hierarchies is definitely interesting.