Sequential Order-Robust Mamba for Time Series Forecasting

Seunghan Lee, Juri Hong, Kibok Lee, Taeyoung Park·October 30, 2024

Summary

The text introduces SOR-Mamba, a time series forecasting method that addresses sequential order bias in multivariate data. It incorporates regularization to minimize distance between embedding vectors generated from data with reversed channel orders, removes a 1D-convolution, and includes Channel Correlation Modeling (CCM) for pretraining. SOR-Mamba demonstrates state-of-the-art performance across standard and transfer learning scenarios on 13 datasets, with greater efficiency compared to previous methods.

Key findings

26

Tables

9

Advanced features