SiloFuse: synthetic data that never leaves the silo

Sharing data between organisations is hard: privacy rules, competitive concerns, and feature partitioning all get in the way. SiloFuse sidesteps the problem by generating synthetic data with latent diffusion models, so each silo keeps its raw records private while still contributing to a shared, realistic dataset.

It is a small example of a larger theme in the lab: getting the value out of data without moving the data itself.