De-Correlation Neural Network for Synchronous Implementation of Estimation and Secrecy

Canyang Guo, Chi-Hua Chen, Ching-Chun Chang, Feng-Jang Hwang, Chin-Chen Chang

Abstract: Intelligent equipment within the Internet of things (IoT) carries out massive, frequent, and persistent data communication, making privacy protection particularly critical. Different from regular encryption methods or neural networks, this study proposes a de-correlation neural network (DeCNN) which synchronously realizes the estimation and privacy protection by a comprehensive loss function. In addition, a two-stage learning algorithm is utilized for solution optimization and computation enhancement. The DeCNN is deployed in the deep fingerprint positioning, and the experimental results demonstrate that the proposed method decreases the maximal correlation coefficient between the transmission data and target data from 0.95 to 0.34 (and 0.13) when the positioning error reaches 1.31 m (and 2.88 m).