Machine learning through cryptographic glasses: combating adversarial attacks by key-based diversified aggregation

摘要

The advent of deep learning techniques [ ] has stimulated the deployment of machine learning in many applications. The DNNs have been applied to solve a wide range of problems in image classification [ , ], object detection [ , ], face recognition [ , ] image caption [ , ], natural language processing [ , ], speech recognition [ , ], drones and robotics [ , ], malware detection [ , ], etc., and more science and discovery-related fields, such as drug composition analysis [ ], brain circuit reconstruction [ ], and DNA mutation impact analysis [ ].