PrivaScissors: Enhance the Privacy of Collaborative Inference through the Lens of Mutual Information
This addresses privacy risks for IoT devices in collaborative inference, though it appears incremental as it builds on prior work to enhance existing defenses.
The paper tackles the problem of data and prediction exposure in edge-cloud collaborative inference by introducing PrivaScissors, a defense strategy that reduces mutual information between intermediate outcomes and device data/predictions, with evaluation on multiple datasets and theoretical robustness guarantees.
Edge-cloud collaborative inference empowers resource-limited IoT devices to support deep learning applications without disclosing their raw data to the cloud server, thus preserving privacy. Nevertheless, prior research has shown that collaborative inference still results in the exposure of data and predictions from edge devices. To enhance the privacy of collaborative inference, we introduce a defense strategy called PrivaScissors, which is designed to reduce the mutual information between a model's intermediate outcomes and the device's data and predictions. We evaluate PrivaScissors's performance on several datasets in the context of diverse attacks and offer a theoretical robustness guarantee.