Please use this identifier to cite or link to this item:
http://artemis.cslab.ece.ntua.gr:8080/jspui/handle/123456789/13472
Full metadata record
DC Field | Value | Language |
---|---|---|
dc.contributor.author | Καραβαλάκης Νικόλαος | |
dc.date.accessioned | 2018-07-23T09:12:44Z | - |
dc.date.available | 2018-07-23T09:12:44Z | - |
dc.date.issued | 2017-7-11 | |
dc.date.submitted | 2017-7-3 | |
dc.identifier.uri | http://artemis-new.cslab.ece.ntua.gr:8080/jspui/handle/123456789/13472 | - |
dc.description.abstract | One of the most essential biological signals for the diagnosis of the heart is the Electrocardiogram (ECG). The need of constant monitoring and on-time heart condition assessment have imposed new requirements for acceleration and power consumption of ECG Analysis Flow. Due to complexity of assessing and predicting heart’s condition, machine learning techniques have become dominant on the field of ECG analysis. Support Vector Machine classifiers is the most efficient way to predict accurately the heart’s condition. Based on multiple computational operations, SVM classifiers lead to excessive power consumption and high execution time. The approach of this thesis is to implement and optimize the algorithm in respect of performance and power consumption. Working towards meeting these specifications, the filter and the Support Vector Machine classifier, the first and the last part of the ECG Analysis Flow respectively, were accelerated, as the most time consuming and power demanding parts of the flow. In this thesis, the ECG Analysis Flow was implemented into an ultra-low-power multicore System-on-Chip Myriad 2, designed for high computational tasks for mobile, wearable and embedded applications, provided by Movidius. Firstly, the original filter and classifier codes are modified, in order to utilize the micro-architectural features and the memory hierarchy of Myriad 2. Afterwards, the computational tasks are delivered to the VLIW micro-processors of Myriad 2, where they are executed in parallel for higher performance efficiency. Finally, our implementation proved that Myriad 2 can achieve up to 97% and 99% latency gain compared to the original filter code and SVM code respectively, coming with significant energy efficiency. | |
dc.language | Greek | |
dc.subject | myriad | |
dc.subject | support vector machine | |
dc.subject | ecg analysis | |
dc.subject | machine learning | |
dc.subject | vectorization | |
dc.title | Αξιολόγησης Της Απόδοσης Και Της Κατανάλωσης Ενεργειας Μίας Εφαρμογής Ανάλυσης Ηλεκτροκαρδιογραφήματος Στη Myriad 2 | |
dc.type | Diploma Thesis | |
dc.description.pages | 93 | |
dc.contributor.supervisor | Σούντρης Δημήτριος | |
dc.department | Τομέας Τεχνολογίας Πληροφορικής & Υπολογιστών | |
dc.organization | ΕΜΠ, Τμήμα Ηλεκτρολόγων Μηχανικών & Μηχανικών Υπολογιστών | |
Appears in Collections: | Διπλωματικές Εργασίες - Theses |
Files in This Item:
File | Size | Format | |
---|---|---|---|
DT2017-0130.pdf | 2.39 MB | Adobe PDF | View/Open |
Items in Artemis are protected by copyright, with all rights reserved, unless otherwise indicated.