On the Role of Reversible Instance Normalization

TL;DR

This paper investigates the role of Reversible Instance Normalization in time series forecasting, revealing that some components are redundant or harmful, and proposes improvements for robustness and generalization.

Contribution

The paper critically analyzes RevIN, identifies redundant components, and offers new perspectives to enhance its robustness and generalization in time series forecasting.

Findings

Some RevIN components are redundant or detrimental.

Revisiting RevIN leads to improved robustness.

Insights contribute to better normalization strategies.

Abstract

Data normalization is a crucial component of deep learning models, yet its role in time series forecasting remains insufficiently understood. In this paper, we identify three central challenges for normalization in time series forecasting: temporal input distribution shift, spatial input distribution shift, and conditional output distribution shift. In this context, we revisit the widely used Reversible Instance Normalization (RevIN), by showing through ablation studies that several of its components are redundant or even detrimental. Based on these observations, we draw new perspectives to improve RevIN's robustness and generalization.