Enhancing privacy in remote data classification

Neural networks are a fundamental tool in data classification. A protocol whereby a user may ask a service provider to ran it neural network on air input provided in encrypted format is proposed here, in such a way that the neural network owner does not get my knowledge about the processed data. At the same time, the knowledge embedded within tire network itself is protected. With respect to previous works in this field, the interaction between the user and the neural network owner is kept to a minimum without resorting to general secure multi-party computation protocols.