Low Latency Privacy Preserving Inference

摘要: When applying machine learning to sensitive data, one has find a balance between accuracy, information security, and computational-complexity. Recent studies combined Homomorphic Encryption with neural networks make inferences while protecting against leakage. However, these methods are limited by the width depth of that can be used (and hence accuracy) exhibit high latency even for relatively simple networks. In this study we provide two solutions address limitations. first solution, present more than 10× improvement in enable inference on wider compared prior attempts same level security. The improved performance is achieved novel represent data during computation. second apply method transfer private services using deep ∼0.16 seconds. We demonstrate efficacy our several computer vision tasks.