Secure distribution of neural networks in wireless sensor networks

Abstract

In this paper we consider the case where artificial neural network processing is securely performed over the wireless sensor network. To do this, we point out major security threats and countermeasures against them. Then, we revised Holenderski et al.'s decomposition model to support secure computing. Moreover, we refine the original model to deal with some boundary cases. The revised model shows that like the original model the horizontal decomposition is better than the vertical decomposition and that the number of the allocated lower neurons in each layer to each sensor node should be large for optimization.