DataMix: Efficient Privacy-Preserving Edge-Cloud Inference

Zhijian Liu, Zhanghao Wu, Chuang Gan, Ligeng Zhu, Song Han ;

Deep neural networks are widely deployed on edge devices (g, for computer vision and speech recognition). Users either perform the inference locally (\ie, edge-based) or send the data to the cloud and run inference remotely (\ie, cloud-based). However, both solutions have their limitations: edge devices are heavily constrained by insufficient hardware resources and cannot afford to run large models; cloud servers, if not trustworthy, will raise serious privacy issues. In this paper, we mediate between the resource-constrained edge devices and the privacy-invasive cloud servers by introducing a novel extit{privacy-preserving edge-cloud inference} framework, \method. We off-load the majority of the computations to the cloud and leverage a pair of mixing and de-mixing operation, inspired by mixup, to protect the privacy of the data transmitted to the cloud. Our framework has three advantages. First, it is extit{privacy-preserving} as the mixing cannot be inverted without the user's private mixing coefficients. Second, our framework is extit{accuracy-preserving} because our framework takes advantage of the space spanned by images, and we train the model in a mixing-aware manner to maintain accuracy. Third, our solution is extit{efficient} on the edge since the majority of the workload is delegated to the cloud, and our mixing and de-mixing processes introduce very few extra computations. Also, our framework introduces small communication overhead and maintains high hardware utilization on the cloud. Extensive experiments on multiple computer vision and speech recognition datasets demonstrate that our framework can greatly reduce the local computations on the edge (to fewer than 20% of FLOPs) with negligible loss of accuracy and no leakages of private information."