Perturbing Inputs to Prevent Model Stealing

Justin Grana

We show how perturbing inputs to machine learning services (ML-service) deployed in the cloud can protect against model stealing attacks. In our formulation, there is an ML-service…