GLAIM: Learning Global and Local Adaptive Inter-Variable Dependency for Multivariate Time Series Imputation

Researchers introduced GLAIM, a new framework for multivariate time series imputation.

It first learns stable global inter-variable dependencies, then refines them per sample and time step using observed values.

The method performed better than prior approaches across nine real-world datasets.

It remained robust under different missing-data patterns and changing missing rates, making it well suited for practical time-series analysis.