<ul><li>Time series imputation is a challenging problem with various applications in fields like health care and the Internet of Things.</li><li>Existing methods often overlook the differences between missing mechanisms (MAR and MNAR) in time series data, leading to misleading results.</li><li>A new framework called Different Missing Mechanisms (DMM) is proposed to address the issue by tailoring solutions based on specific missing mechanisms.</li><li>The method utilizes variational inference and normalizing flow-based neural architecture to model data generation processes, showing improved performance over existing techniques in real-world applications.</li></ul>

Causal View of Time Series Imputation: Some Identification Results on Missing Mechanism

Discover more