ICML 2022 ぷ the latest fedformer of the Dharma Institute of Afghanistan ⻓ surpasses SOTA in the whole process of time series prediction

Recently, the decision-making Intelligence Laboratory of Alibaba Dharma Academy was in ICML 2022 The latest work in the direction of time series prediction was published on ： Frequency domain decomposition based FEDformer Model . In the long time series prediction problem , Computational efficiency 、 The prediction accuracy has been greatly improved .

Thesis link ： https://arxiv.org/abs/2201.12740

Code link ： https://github.com/DAMO-DI-ML/ICML2022-FEDformer

The author of this paper puts forward FEDformer By using the low-rank approximation It can achieve linear complexity , And greatly surpass in accuracy SOTA（state-of-the-art） result .