Predicting Failures in Embedded Systems Using Long Short-Term Inference

Users of embedded and cyber-physical systems expect dependable operation for an increasingly diverse set of applications and environments. Reactive self-diagnosis techniques either use unnecessarily conservative guardbands, or do not prevent catastrophic failures. In this letter, we utilize machine-learning techniques to design a prediction engine in order to predict failures on-device in the embe...