Quantization with Gate Disclosure for Embedded Artificial Intelligence Applied to Fall Detection
Sergio Duarte Correia, João Pedro Matos-Carvalho, Slaviša Tomic · 2024
Fall detection in the elderly population is a typical application of pattern recognition, where machine learning algorithms have shown good performance results in the scientific literature. Nevertheless, the usual large dimension of the networks proposes a challenge for embedded implementations that could be used on wearable or wireless sensor networks. The most common implementation relies on edge or cloud computing of the algorithms, which triggers potential privacy issues and poses a challenge in terms of a large amount of data that need to be transferred through a communication channel. The current work proposes a new methodology for performing network quantization. The extensive simulation results demonstrate the effectiveness and feasibility of the employed methodology for embedded implementation of the LSTM for the fall detection problem on wearable platforms.