Sequential Order-Robust Mamba for Time Series Forecasting

TL;DR

The paper introduces SOR-Mamba, a novel time series forecasting method that improves robustness to channel order variations and enhances channel dependency modeling by combining regularization and correlation preservation techniques.

Contribution

It proposes SOR-Mamba, which removes order bias and incorporates channel correlation modeling, advancing time series forecasting with improved robustness and dependency capture.

Findings

Enhanced robustness to channel order variations

Effective preservation of channel correlations

Improved forecasting accuracy in experiments

Abstract

Mamba has recently emerged as a promising alternative to Transformers, offering near-linear complexity in processing sequential data. However, while channels in time series (TS) data have no specific order in general, recent studies have adopted Mamba to capture channel dependencies (CD) in TS, introducing a sequential order bias. To address this issue, we propose SOR-Mamba, a TS forecasting method that 1) incorporates a regularization strategy to minimize the discrepancy between two embedding vectors generated from data with reversed channel orders, thereby enhancing robustness to channel order, and 2) eliminates the 1D-convolution originally designed to capture local information in sequential data. Furthermore, we introduce channel correlation modeling (CCM), a pretraining task aimed at preserving correlations between channels from the data space to the latent space in order to enhance the ability to capture CD. Extensive experiments demonstrate the efficacy of the proposed method across standard and transfer learning scenarios. Code is available at https://github.com/seunghan96/SOR-Mamba.