Resources
Notes from the work, numbers included.
Technical write-ups on private synthetic clinical data: how we measure it, where it works, and where it still falls short. Written for biostatisticians, data managers and privacy officers.
Benchmark
Explainer
Methods note
Private synthetic data on five public clinical datasets: what held up and what didn't
Fidelity, predictive utility and privacy-attack results at ε 8, with the baselines that put each number in context and the places the engine falls short.
9 min readWhat ε = 8 means, and what it doesn't: the privacy budget in plain language
What the privacy budget actually guarantees, why ε = 8 is less alarming than its worst-case number, where privacy usually leaks, and five questions to ask any vendor.
7 min readSynthetic data for small cohorts: what works with 300 patients
Why private neural generators fail on a few hundred patients, what a private copula recovers, and the limits that matter for rare-disease and early-phase teams.
6 min read