A new math for combining data from different sources without forcing them into the same shape

Researchers developed a way to merge multiple datasets that measure the same thing differently — without flattening them into a single format first. Instead of concatenating all the data together or assuming everything fits one geometry, the new method uses optimal transport math to find a common structure that respects what makes each view distinct.

Most data fusion methods treat all datasets the same way: they either stack them side by side or assume they can be warped into alignment. This breaks when the datasets have fundamentally different shapes — a medical scan and a genetic profile aren't just rotations of each other. The new approach solves a harder problem: finding the coherent low-dimensional structure without forcing the data to conform to a single geometry. This matters for any domain where you're combining heterogeneous data sources and don't want to lose the structure that makes each one informative.