SeesawNet: Towards Non-stationary Time Series Forecasting with Balanced Modeling of Common and Specific Dependencies

TL;DR

SeesawNet is a novel architecture for non-stationary time series forecasting that adaptively balances modeling common and specific dependencies across temporal and channel dimensions.

Contribution

It introduces Adaptive Stationary-Nonstationary Attention (ASNA) to dynamically fuse shared and instance-specific dependencies, improving forecasting accuracy.

Findings

SeesawNet outperforms existing methods on multiple benchmarks.

ASNA effectively captures both common and specific dependencies.

The model adapts to non-stationary structures in time series data.

Abstract

Instance normalization (IN) is widely used in non-stationary multivariate time series forecasting to reduce distribution shifts and highlight common patterns across samples. However, IN can over-smooth instance-specific structural information that is essential for modeling temporal and cross-channel heterogeneity. While prior methods further suppress distribution discrepancies or attempt to recover temporal specific dependencies, they often ignore a central tension: how to adaptively model common and instance-specific dependency based on each instance's non-stationary structures. To address this dilemma, we propose SeesawNet, a unified architecture that dynamically balances common and instance-specific dependency modeling in both temporal and channel dimensions. At its core is Adaptive Stationary-Nonstationary Attention (ASNA), which captures common dependencies from normalized sequences and specific dependencies from raw sequences, and adaptively fuses them according to instance-level non-stationarity. Built upon ASNA, SeesawNet alternates dedicated temporal and channel relationship modeling to jointly capture long-range and cross-variable dependencies. Extensive experiments on multiple real-world benchmarks demonstrate that SeesawNet consistently outperforms state-of-the-art methods.