Rethinking Time Series Domain Generalization via Structure-Stratified Calibration

2026-03-03

AI Summary

This work addresses the challenge of cross-domain generalization in time series by revealing that existing methods often suffer from spurious alignment and negative transfer due to neglecting structural discrepancies between dynamic systems. The study reframes the problem through the lens of structural correspondence failure and introduces a novel paradigm: enforcing structural consistency prior to distributional alignment. To this end, the authors propose a structure-aware hierarchical calibration framework that first identifies cross-domain sample clusters with consistent underlying dynamics via structure-aware clustering, then performs amplitude calibration within each cluster. This approach effectively circumvents generalization failures caused by structural incompatibility. Extensive experiments across 19 public datasets (comprising 100.3k samples) demonstrate that the method significantly outperforms strong baselines, underscoring the critical role of structural consistency in enhancing zero-shot cross-domain generalization performance.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Application Domains: Humanities & Computational Social Science

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Bridging structured and unstructured data

Abstract

For time series arising from latent dynamical systems, existing cross-domain generalization methods commonly assume that samples are comparably meaningful within a shared representation space. In real-world settings, however, different datasets often originate from structurally heterogeneous families of dynamical systems, leading to fundamentally distinct feature distributions. Under such circumstances, performing global alignment while neglecting structural differences is highly prone to establishing spurious correspondences and inducing negative transfer. From the new perspective of cross-domain structural correspondence failure, we revisit this problem and propose a structurally stratified calibration framework. This approach explicitly distinguishes structurally consistent samples and performs amplitude calibration exclusively within structurally compatible sample clusters, thereby effectively alleviating generalization failures caused by structural incompatibility. Notably, the proposed framework achieves substantial performance improvements through a concise and computationally efficient calibration strategy. Evaluations on 19 public datasets (100.3k samples) demonstrate that SSCF significantly outperforms strong baselines under the zero-shot setting. These results confirm that establishing structural consistency prior to alignment constitutes a more reliable and effective pathway for improving cross-domain generalization of time series governed by latent dynamical systems.

Problem

• time series
• domain generalization
• dynamical systems
• structural heterogeneity
• negative transfer

Innovation

• structure-stratified calibration
• time series domain generalization
• latent dynamical systems
• structural consistency
• negative transfer