Nature Communications · 19 September 2026 · From issue 39
Dynamics-informed machine learning for recovering extensive missing systems dynamics
A computer model rebuilt missing data using shared patterns, even when 90% was gone.
Tao Wu, Xiangyun Gao et al.
A new computer method fills in missing data in complex time series. It looks at how one measurement's pattern moves over time and matches that shape to another measurement's shape. This lets it fill in gaps even when most of the data is missing. It worked on made-up test systems and real data like traffic, brain signals, money exchange rates, and wind speed.
The amount of missing data the new method could still recover, if one variable stayed complete.
Tested on simulations and a few real datasets, not all data types.
Integrity screen: passed (3 checks) Checked 22 September 2026. Retraction record: none. DOI resolves at doi.org. Metadata record found (Nature Communications). Read the source
This screen checks process signals. It does not tell you the findings are true.