Reproduce “S4M: S4 for multivariate time series forecasting with Missing values” 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
Jing Peng, Meiqi Yang, Qiong Zhang, Xiaoxiao Li
The S4M framework, developed by researchers from Renmin University of China, Princeton University, the University of British Columbia, and the Vector Institute, integrates missing data handling directly into multivariate time series forecasting, extending the Structured State Space Sequence (S4) model. It consistently outperforms various baseline methods on real-world datasets, achieving lower forecasting errors while maintaining computational efficiency.
Key insights
- Integrating missing data handling directly into the forecasting model avoids error propagation inherent in separate imputation and forecasting steps.
- Leveraging historical patterns through a prototype bank can robustly represent and account for missing values during forecasting.
- Explicitly incorporating missingness masks into state-space equations allows the model to adapt its learning and predictions based on data availability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.