Reproduce “Merlin: Multi-View Representation Learning for Robust Multivariate Time Series Forecasting with Unfixed Missing Rates” 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
Chengqing Yu, Fei Wang, Chuanguang Yang, Zezhi Shao, Tao Sun, Tangwen Qian, Wei Wei, Zhulin An and 1 more
Merlin trains a forecasting student across randomly masked sensor histories, using a teacher trained on complete data to align hidden representations and forecasts, alongside contrastive learning across missing rates. In a simulated changing-rate test, its PEMS04 MAE was 21.18, versus 22.15 for rate-specific STID+GATGPT. The test covered METR-LA and PEMS04, so the result does not establish robustness to other failure patterns.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.