Abstract | ||
---|---|---|
Neural network is a particular machine learning framework which has gained widespread popularity due to its superior performance in many applications, such as complex board games, face recognition and disease diagnosis. A new service paradigm which offers online neural-network-based prediction to clients is increasingly popular. Although the prediction service has clear benefits, serious privacy issues have also emerged from the clients’ sensitive data and the neural network model itself. In this paper, we present a new outsourcing model for privacy-preserving neural network prediction under two non-colluding servers framework. In this model, the original neural network owner can securely outsource an existing neural network model to the two servers who will thereafter provide prediction service to the public users utilize encrypted input data. We propose the first fully non-interactive privacy-preserving neural network prediction scheme. Extensive security and efficiency analysis demonstrate that the proposed scheme satisfies the security requirement of model privacy and data privacy, and highly efficient with respect to computation and communication overhead. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1016/j.ins.2018.12.015 | Information Sciences |
Keywords | Field | DocType |
Outsourcing computation,Homomorphic encryption,Neural network prediction,Privacy preservation | Facial recognition system,Server,Popularity,Outsourcing,Encryption,Artificial intelligence,Information privacy,Artificial neural network,Mathematics,Machine learning,Computation | Journal |
Volume | ISSN | Citations |
481 | 0020-0255 | 5 |
PageRank | References | Authors |
0.43 | 28 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Xu Ma | 1 | 21 | 4.12 |
Xiaofeng Chen | 2 | 2603 | 141.37 |
Xiaoyu Zhang | 3 | 112 | 23.80 |