toy time series dataset
收藏资源简介:
本文设计了一个玩具时间序列数据集,用于研究注意力机制在时间序列预测中的表现。该数据集旨在揭示注意力机制在时间序列预测中的失败原因,并探讨其背后的理论依据。数据集的创建过程包括设计不同的注意力机制和模型架构,并通过实验验证了注意力机制在时间序列预测中的有效性。数据集的应用领域主要包括时间序列预测,旨在解决时间序列Transformer模型中注意力机制失败的问题。
This paper proposes a toy time series dataset for studying the performance of attention mechanisms in time series forecasting. This dataset is designed to uncover the reasons behind the failure of attention mechanisms in time series forecasting and explore their underlying theoretical foundations. The development of this dataset includes designing various attention mechanisms and model architectures, and validating the effectiveness of attention mechanisms in time series forecasting via experiments. The primary application scope of this dataset is time series forecasting, with the goal of addressing the failure problem of attention mechanisms in time series Transformer models.

- 1通过上海交通大学 · 2025年



