Today differential privacy has been used widely to prevent data reconstruction attacks? There are several algorithms in the literature that can construct original data from the gradients obtained during the training process of distributed machine learning. When we try to reconstruct the data with these differentially private gradients, we get very poor accuracy in building the original data. Have anyone tried to develop an algorithm that can extract the necessary knowledge from the differentially private gradients achieving a significant amount of accuracy(>90%) in data reconstruction?
Rajat Tandon
Not sure if it is very relevant, but we did something to improve geo-location noise.
https://link.springer.com/chapter/10.1007%2F978-3-030-03405-4_5
0 votes
0 thanks
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts
More Krishna Yadav's questions See All
Similar questions and discussions