《A DECODER-ONLY FOUNDATION MODEL FOR TIME-SERIES FORECASTING》阅读总结

介绍了一个名为TimeFM的新型时间序列预测基础模型，该模型受启发于自然语言处理领域的大语言模型，通过再大规模真实世界和合成时间序列数据集上的预训练，能够在多种不同的公共数据集上实现接近最先进监督模型的零样本预测性能。

该模型使用真实世界和合成数据集构建的大型时间序列语料库进行预训练，并展示了在不同领域、预测范围和时间粒度的未见数据集上的准确零样本预测能力。

1、引言

时间序列在零售、金融、制造业、医疗保健和自然科学等各个领域无处不在。近年来，深度学习模型已成为预测丰富多元时间序列的流行方法。

深度学习模型： 

1、David Salinas, V alentin Flunkert, Jan Gasthaus, and Tim Januschowski. Deepar: Probabilistic forecasting with autoregressive recurrent networks. International Journal of F orecasting, 36(3):1181–1191, 2020.

2、Boris N Oreshkin, Dmitri Carpov, Nicolas Chapados, and Y oshua Bengio. N-beats: Neural basis expansion analysis for interpretable time series forecasting. In International Conference on Learning Representations, 2019.

3、Rajat Sen, Hsiang-Fu Y u, and Inderjit S Dhillon. Think globally, act locally: A deep neural network approach to high-dimensional time series forecasting. Advances in neural information processing systems, 32, 2019.

4、Tian Zhou, Ziqing Ma, Qingsong Wen, Xue Wang, Liang Sun, and Rong Jin. Fedformer: Frequency enhanced decomposed transformer for long-term series forecasting. In International Conference on Machine Learning, pages 27268–27286. PMLR, 2022.

5、Si-An Chen, Chun-Liang Li, Nate Y oder, Sercan O Arik, and Tomas Pfister. Tsmixer: An all-mlp architecture fo