no code implementations • 8 Sep 2023 • Daniel Scheliga, Patrick Mäder, Marco Seeland
To preserve the privacy preserving effect of PRECODE, our analysis reveals that variational modeling must be placed early in the network.
1 code implementation • 12 Aug 2022 • Daniel Scheliga, Patrick Mäder, Marco Seeland
We find that state of the art attacks are not able to reconstruct the client data due to the stochasticity induced by dropout during model training.
no code implementations • 9 Aug 2022 • Daniel Scheliga, Patrick Mäder, Marco Seeland
In result, we show that our approach requires less gradient perturbation to effectively preserve privacy without harming model performance.
1 code implementation • 10 Aug 2021 • Daniel Scheliga, Patrick Mäder, Marco Seeland
We propose a simple yet effective realization of PRECODE using variational modeling.