MtsCID: A Dual‑Network Framework for Coarse‑Grained Multivariate Time Series Anomaly Detection

We present *MtsCID*, a semi‑supervised anomaly‑detection framework for multivariate time series that jointly learns coarse‑grained temporal and inter‑variates dependencies in time and frequency domains. Building on dual‑branch architecture and prototype‑guided attention, MtsCID achieves state‑of‑the‑art performance on seven benchmark datasets while remaining computationally efficient.