Reproduce “GLA-DA: Global-Local Alignment Domain Adaptation for Multivariate Time Series” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Gang Tu, Dan Li, Bingxin Lin, Zibin Zheng, See-Kiong Ng
GLA-DA, developed by researchers at Sun Yat-Sen University and NUS, introduces a domain adaptation framework for multivariate time series that integrates global and local feature alignment. This method achieved superior Macro-F1 scores, outperforming state-of-the-art baselines in 17 out of 20 evaluated scenarios for human activity recognition and sleep stage classification.
Key insights
- Effective domain adaptation for time series requires not only global feature alignment but also explicit local class alignment to maintain class discriminability.
- A robust pseudo-labeling mechanism, such as the proposed "Agree Mechanism" that leverages consistency between different models, is crucial for effectively utilizing unlabeled target data in unsupervised and semi-supervised settings.
- Traditional vision-centric backbone architectures are suboptimal for time series data; 1D-CNNs are better suited to capture the unique temporal characteristics.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.