Differential Privacy Has Disparate Impact on Model Accuracy

Eugene Bagdasaryan
Eugene Bagdasaryan
Omid Poursaeed
Omid Poursaeed

ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 32 (NIPS 2019), pp. 15453-15462, 2019.

Cited by: 17|Bibtex|Views28
EI
Other Links: academic.microsoft.com|dblp.uni-trier.de

Abstract:

Differential privacy (DP) is a popular mechanism for training machine learning models with bounded leakage about the presence of specific points in the training data. The cost of differential privacy is a reduction in the model's accuracy. We demonstrate that in the neural networks trained using differentially private stochastic gradient ...More

Code:

Data:

Full Text
Your rating :
0

 

Tags
Comments