Edge Acceleration for Machine Learning-based Motion Artifact Detection on fNIRS Dataset

Abstract

Machine Learning has potential applications across a wide spectrum of devices. However, current approaches for domain-specific accelerators have encountered difficulties in satisfying the most recent computational demands for machine learning applications. This work aims to create an adaptive acceleration framework for fNIRS motion artefact detection, which will be specifically designed for wearable devices. We evaluate the performance of the SVM classifier that has been implemented using SYCL on our fNIRS dataset across diverse devices and discuss the potential to accelerate more advanced motion artefact classifiers at the edge.