Discretizing Continuous Time Series for Imputation with Masked Diffusion Training
A smarter way to fill in missing data points in time series
Researchers developed a new method called MDTIM that fills in missing values in time series data—like temperature readings with gaps or stock prices with incomplete records—by treating masked (missing) and observed values as fundamentally different things during training. Rather than training on noise prediction like existing approaches, the model learns to directly predict the actual missing values, and a new discretization technique lets it handle continuous data while maintaining awareness of the ordering between values.
Time series data with gaps are everywhere: weather stations with sensor failures, medical monitoring with dropped readings, industrial equipment with interrupted logging. Better imputation means more reliable downstream analysis and forecasting, reducing errors in climate models, patient diagnostics, and predictive maintenance. This approach outperformed existing methods across different types and amounts of missing data, suggesting it could become a standard tool for cleaning real-world time series in practice.