PSformer: Parameter-efficient Transformer with Segment Shared Attention for Time Series Forecasting

TMLR Paper4688 Authors

Abstract: Time series forecasting remains a critical challenge across various domains, often complicated by high-dimensional data and long-term dependencies. This paper presents a novel transformer architecture for time series forecasting, incorporating two innovative designs: parameter sharing module (PS) and Segment Shared Attention (SSA). The proposed model, PSformer, reduces the number of training parameters through the integrated parameter sharing mechanism without sacrificing performance. The spatiotemporal segment defined as a patch spanning across spatial variables and local time. The introduction of SSA could enhance the capability of capturing local spatio-temporal dependencies and improve global representation by integrating information across segments. Consequently, The combination of parameter sharing and SSA reduces the model's parameter count while enhancing forecasting performance. Extensive experiments on benchmark datasets demonstrate that PSformer outperforms many baseline approaches in terms of accuracy and scalability, positioning it as an effective and scalable tool for time series forecasting. Code can be found in https://anonymous.4open.science/r/PSformer_Anonymous-3E11.

Submission Length: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Min_Wu2

Submission Number: 4688