Reproduce “TimesURL: Self-supervised Contrastive Learning for Universal Time Series Representation Learning” 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
Jiexi Liu, Songcan Chen
TimesURL, a novel self-supervised learning framework, generates high-quality, universal time series representations by integrating a frequency-temporal augmentation strategy, a double Universum hard negative sampling technique, and a joint contrastive and reconstruction loss. This framework consistently achieved leading performance across six diverse downstream tasks including forecasting, classification, and anomaly detection.
Key insights
- Effective representation learning for time series benefits from augmentation strategies that carefully preserve inherent temporal dynamics and semantic consistency.
- Systematically generating hard negative samples, such as through Double Universums, significantly enhances the discriminative power and generalizability of contrastive learning for time series.
- A comprehensive time series representation requires jointly capturing both fine-grained, segment-level temporal dynamics and coarse-grained, instance-level global contextual information.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.