Capturing The Channel Dependency Completely Via Knowledge-Episodic Memory For Time Series Forecasting

Muyao Wang, Wenchao Chen, Zhibin Duan, Binhe Zhao, Bo Chen

Primary Area: representation learning for computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: multivariate time series forecasting;channel dependency;pattern memory network;

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: Using pattern memory network to capture the channel dependency for multivariate time series forecasting.

Abstract:

The forecasting of Multivariate Time Series (MTS) has long been an important but challenging task, and recent advancements in MTS forecasting methods try to discover both temporal and channel-wise dependencies. However, we explore the nature of MTS and observe two kinds of existed channel dependencies that current methods have difficulty to capture completely.
One is the evident channel dependency, which can be captured by mixing the channel information directly, and another is the latent channel dependency, which should be captured by finding the intrinsic variable that caused the same changes within MTS.
To address this issue, we introduce the knowledge and episodic memory modules, which gain the specific knowledge and hard pattern memories with a well-designed recall method, to capture the latent and evident channel dependency respectively.
Further, based on the proposed memory modules, we develop a pattern memory network, which recalls both memories for capturing different channel dependencies completely, for MTS forecasting. Extensive experiments on eight datasets all verify the effectiveness of the proposed memory-based forecasting method.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 7117