Phase-driven Domain Generalizable Learning for Nonstationary Time Series

2024-02-05

arXiv.org

AI Summary

To address insufficient cross-domain generalization of non-stationary time series under continuous sensing due to distributional shift, this paper proposes PhASER. The method first uncovers the intrinsic relationship between phase information and non-stationary statistical properties, and innovatively introduces phase enhancement, amplitude-phase dual-modality disentangled encoding, and a phase-guided residual feature broadcasting mechanism. Leveraging STFT-based time-frequency modeling, phase-sensitive data augmentation, and domain-invariant representation learning, PhASER significantly improves model robustness against dynamic distribution shifts. Evaluated on five real-world non-stationary time-series datasets, PhASER achieves an average accuracy gain of 5.0% over ten state-of-the-art methods, with a maximum improvement of 13%. Moreover, it serves as a plug-and-play module that enhances the generalization capability of existing classifiers without architectural modification.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Multi-modal Vision

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

Monitoring and recognizing patterns in continuous sensing data is crucial for many practical applications. These real-world time-series data are often nonstationary, characterized by varying statistical and spectral properties over time. This poses a significant challenge in developing learning models that can effectively generalize across different distributions. In this work, based on our observation that nonstationary statistics are intrinsically linked to the phase information, we propose a time-series learning framework, PhASER. It consists of three novel elements: 1) phase augmentation that diversifies non-stationarity while preserving discriminatory semantics, 2) separate feature encoding by viewing time-varying magnitude and phase as independent modalities, and 3) feature broadcasting by incorporating phase with a novel residual connection for inherent regularization to enhance distribution invariant learning. Upon extensive evaluation on 5 datasets from human activity recognition, sleep-stage classification, and gesture recognition against 10 state-of-the-art baseline methods, we demonstrate that PhASER consistently outperforms the best baselines by an average of 5% and up to 13% in some cases. Moreover, PhASER's principles can be applied broadly to boost the generalization ability of existing time series classification models.

Problem

• Addressing distribution shifts in nonstationary time series classification
• Learning generalizable representations using phase information as proxy
• Developing phase-driven framework to handle statistical property variations

Innovation

• Hilbert transform-based augmentation diversifies nonstationarity
• Separate magnitude-phase encoding treats them as modalities
• Phase-residual feature broadcasting integrates 2D phase features