Recent hardware and software advances allow efficient local machine learning inference on edge devices such as smart cameras. This feature addresses users’ privacy concerns, as they no longer need to upload sensitive raw data, such as images and videos, to cloud servers for data analysis. This paper studies potential privacy breaches in this emerging computing paradigm. Specifically, we show that it is possible to manipulate production neural network models to turn them into spyware. The inference results of these models are almost the same as they are expected to be, but can convey information about the raw input content, allowing the curious cloud service provider managing the edge devices to covertly spy on the private information without being noticed. In the end, we discuss possible defense approaches.

Emerging mobile applications involving distributed control and sensing call for accurate time synchronization over wireless links. This paper systematically studies the performance of Precision Time Protocol (PTP) in today’s Wi-Fi networks. We investigate both software and hardware PTP implementations. Our study uncovers the root causes of software PTP synchronization errors. We show that with fine-tuned system configurations and an online calibration procedure, software PTP can achieve reasonable accuracy with off-the-shelf Wi-Fi devices. Hardware PTP requires a PTP hardware timestamper clock not contained in Wi-Fi NICs. We propose a method to make use of the hardware TSF counter to emulate the PTP clock. Rigorous tests traversing various conditions show that both software and hardware PTP implementations can achieve 1-$\mu s$level accuracy on current ARM and x86 mobile platforms with practical Wi-Fi settings.