Privacy Amplification Persists under Unlimited Synthetic Data Release
Clément Pierquin, Aurélien Bellet, Marc Tommasi, Matthieu Boussard · 2026
Read Privacy Amplification Persists under Unlimited Synthetic Data Release by Clément Pierquin, Aurélien Bellet, Marc Tommasi, Matthieu Boussard (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-04895.txt.
Description
Synthetic data, fake data generated to stand in for real records, can strengthen privacy guarantees when it is released instead of the model trained on private data. The paper comes from Clément Pierquin and 3 co-authors. Earlier work proved this effect only when the number of released records stayed small compared with the size of the model.
The team shows the effect does not vanish as more records are released. Under a bound on the model's parameters, the privacy boost survives even when an unlimited number of synthetic records is published. Their proof improves on the earlier bounds and explains the structure behind the effect. The authors suggest the insight could lead to tighter privacy guarantees for more complex systems.
Catalog record
| CREATOR | Clément Pierquin, Aurélien Bellet, Marc Tommasi, Matthieu Boussard |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 133 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-04895