Alexander Benvenuti

End-to-End Differential Privacy in Training Deep Neural Network Classifiers

Huaiyuan Rao, Calvin Hawkins, Alexander Benvenuti, Matthew Hale

Differentially private machine learning enables model training on sensitive data while ensuring that individual data is unlikely to be recoverable from the parameters of the resulting model. However, existing work often privatizes both training inputs and their labels, and these…Read more