报告题目：Crossformer: Transformer Utilizing Cross-Dimension Dependency for Multivariate Time Series Forecasting

论文出处：ICLR2023

作者：Yunhao Zhang & Junchi Yan单位：MoE Key Lab of Artificial Intelligence, Shanghai Jiao Tong University and Shanghai AI Lab

报告人：康汝兵

报告时间：2024.5.20

报告地点：博学楼621

报告内容摘要：Recently many deep models have been proposed for multivariate time series (MTS)forecasting. In particular, Transformer-based models have shown great potentialbecause they can capture long-term dependency. However, existing Transformerbased models mainly focus on modeling the temporal dependency (cross-timedependency) yet often omit the dependency among different variables (crossdimension dependency), which is critical for MTS forecasting. To fill the gap, wepropose Crossformer, a Transformer-based model utilizing cross-dimension dependency for MTS forecasting. In Crossformer, the input MTS is embedded into a 2Dvector array through the Dimension-Segment-Wise (DSW) embedding to preservetime and dimension information. Then the Two-Stage Attention (TSA) layer isproposed to efficiently capture the cross-time and cross-dimension dependency.Utilizing DSW embedding and TSA layer, Crossformer establishes a HierarchicalEncoder-Decoder (HED) to use the information at different scales for the finalforecasting. Extensive experimental results on six real-world datasets show theeffectiveness of Crossformer against previous state-of-the-arts.