13 December 2017

Private Machine Learning

@ mortendahi Encrypted data set (masked images) => transform encrypted data (transform masked values/ arithmetic operations) => encrypted model (masked weights)

data keys(masks) => transform keys => model keys

=> model prediction

OpenMined project

Q: Private ML makes it harder for ML interpretability ? A: Train on public data to understand, then transfered learning on private data.



blog comments powered by Disqus

Number of visits: - |