Abstract
The aim of this dissertation is to explore Human Machine Interfaces (HMIs) in a variety of biomedical scenarios. The research addresses typical challenges in wearable and implantable devices for diagnostic, monitoring, and prosthetic purposes, suggesting a methodology for tailoring such applications to cutting edge embedded architectures. The main challenge is the enhancement of high-level applications, also introducing Machine Learning (ML) algorithms, using parallel programming and specialized hardware to improve the performance. The majority of these algorithms are computationally intensive, posing significant challenges for the deployment on embedded devices, which have several limitations in term of memory size, maximum operative frequency, and battery duration. The proposed solutions take advantage of a Parallel Ultra-Low Power (PULP) architecture, enhancing the elaboration on specific target architectures, heavily optimizing the execution, exploiting software and hardware resources. The thesis starts by describing a methodology that can be considered a guideline to efficiently implement algorithms on embedded architectures. This is followed by several case studies in the biomedical field, starting with the analysis of a Hand Gesture Recognition, based on the Hyperdimensional Computing algorithm, which allows performing a fast on-chip re-training, and a comparison with the state-of-the-art Support Vector Machine (SVM); then a Brain Machine Interface (BCI) to detect the respond of the brain to a visual stimulus follows in the manuscript. Furthermore, a seizure detection application is also presented, exploring different solutions for the dimensionality reduction of the input signals. The last part is dedicated to an exploration of typical modules for the development of optimized ECG-based applications.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have