Not Just Privacy: Improving Performance of Private Deep Learning in Mobile Cloud

The increasing demand for on-device deep learning services calls for a highly manner to deploy deep neural networks (DNNs) on mobile devices with capacity. The cloud-based solution is a promising approach to enabling learning applications on mobile devices where the large portions of a DNN offloaded to the cloud. However, revealing data to the cloud leads to privacy risk. To benefit from the cloud data center without the risk, we design, evaluate, and implement a cloud-based framework ARDEN partitions the DNN across mobile devices and cloud data centers. A simple transformation is performed on the mobile device, while the-hungry training and the complex inference rely on the cloud data. To protect the sensitive information, a lightweight privacy-preserving consisting of arbitrary data nullification and random noise addition introduced, which provides strong privacy guarantee. A rigorous privacy analysis is given. Nonetheless, the private perturbation to the original inevitably has a negative impact on the performance of further inference the cloud side. To mitigate this influence, we propose a noisy training to enhance the cloud-side network robustness to perturbed data. Through sophisticated design, ARDEN can not only preserve privacy but also improve inference performance. To validate the proposed ARDEN, a series of based on three image datasets and a real mobile application are. The experimental results demonstrate the effectiveness of ARDEN. , we implement ARDEN on a demo system to verify its practicality.