Search Results (3)

In this paper, we introduce a security approach for on-device learning Edge AIs designed to detect abnormal conditions in factory machines. Since Edge AIs are easily accessible by an attacker physically, there are security risks due to physical attacks. In particular, there is a concern that the attacker may tamper with the training data of the on-device learning Edge AIs to degrade the task accuracy. Few risk assessments have been reported. It is important to understand these security risks before considering countermeasures. In this paper, we demonstrate a data poisoning attack against an on-device learning Edge AI. Our attack target is an on-device learning anomaly detection system. The system adopts MEMS accelerometers to measure the vibration of factory machines and detect anomalies. The anomaly detector also adopts a concept drift detection algorithm and multiple models to accommodate multiple normal patterns. For the attack, we used a method in which measurements are tampered with by exposing the MEMS accelerometer to acoustic waves of a specific frequency. The acceleration data falsified by this method were trained on an anomaly detector, and the result was that the abnormal state could not be detected. Full article

Background and Objectives: The efficiency and optimal voice rest period following phonosurgery remains debatable. Platelet-rich plasma (PRP) is a safe and cheap alternative to many bioactive agents being studied on animal models, and is already in use in many medical areas. We investigate the short-term effects of PRP and voice rest on voice outcomes following phonosurgery as an alternative to voice rest alone. Materials and Methods: A prospective single-blinded pilot study was conducted. Sixteen patients with a diagnosis of vocal fold cyst and polyps were included, forming equal groups (PRP and voice rest vs. voice rest alone). Voice analysis was carried out on the preoperative day, day three, and week three following surgery. The measured parameters were fundamental frequency (F0), noise–signal ratio (NSR), harmonic poverty (HP), attack alteration (AL), pitch instability (PI), and amplitude instability (AI).VHI(Voice Handicap Index)-30 questionnaires were carried out before surgery and three weeks following surgery to assess the impact of subjective voice change on quality of life. PRP was obtained using commercial kits with separator gel. Results: An average 3.68-fold increase in platelets was obtained with PRP. No side effects were noted after injection. All voice parameters improved on day three and week three following surgery. Statistical significance was noted only in the fundamental frequency of male patients (p = 0.048) in favor of the PRP-voice rest group. In addition, the VHI- 30 questionnaire results between preoperative and postoperative assessments showed statistically significant differences in total VHI score (p = 0.02) as well as the physical (p = 0.05) and emotional (p = 0.02) scale in favor of the PRP-voice rest group. Conclusions: PRP presents short term safety in patients who undergo phonosurgery, although long-term outcomes are unknown. PRP and voice rest are superior to voice rest alone when considering subjective assessment of the voice. When analyzing acoustic parameters, PRP and voice rest are not superior to voice rest alone. Full article

In order to quantitatively study the interfered output of the accelerometer under an acoustic injection attack, a mathematical model for fitting and predicting the accelerometer output was proposed. With ADXL103 as an example, an acoustic injection attack experiment with amplitude sweeping and frequency sweeping was performed. In the mathematical model, the R-squared coefficient was R² = 0.9990 in the acoustic injection attack experiment with amplitude sweeping, and R² = 0.9888 with frequency sweeping. Based on the mathematical model, the dual frequency acoustic injection attack mode was proposed. The difference frequency signal caused by the nonlinear effect was not filtered by the low-pass filter. At a 115 dB sound pressure level, the maximum acceleration bias of the output was 4.4 m/s² and the maximum amplitude of fluctuation was 4.97 m/s². Two kinds of methods of prevention against acoustic injection attack were proposed, including changing the damping ratio of the accelerometer and adding a preposition low-pass filter. Full article