Συντάχθηκε 13-10-2022 08:47
Τόπος:
Σύνδεσμος τηλεδιάσκεψης
Έναρξη: 14/10/2022 15:00
Λήξη: 14/10/2022 16:00
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ
Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών
Πρόγραμμα Προπτυχιακών Σπουδών
ΠΑΡΟΥΣΙΑΣΗ ΔΙΠΛΩΜΑΤΙΚΗΣ ΕΡΓΑΣΙΑΣ
ΑΘΑΝΑΣΙΟΥ ΡΕΝΤΖΕΠΟΠΟΥΛΟΥ
με θέμα
Πλατφόρμα αξιολόγησης αλγορίθμων συμπερασμού με δεδομένα από δίκτυα αισθητήρων
A platform to benchmark inference algorithms based on sensor network data
Εξεταστική Επιτροπή
Καθηγητής Άγγελος Μπλέτσας (επιβλέπων)
Αναπληρωτής Καθηγητής Βασίλειος Σαμολαδάς
Καθηγητής Γεώργιος Καρυστινός
Abstract
This work offers an online tool to perform inference on sensor network (e.g. soil moisture) measurements/data. The design problem is approached by creating a web-based user interface (UI) application based on microservices with a scalable back-end that runs the algorithms and a database storage. Algorithms are expected to be Python scripts developed offline and tested either offline or online. This multi-modal operation poses a compatibility concern. We create a library that exclusively handles data input and output for the scripts that import it. We create different variations of the library depending on the context of the user scripts. We also explore a powerful way of executing scripts, in the back-end of the web app, allowing for robust interruption of running scripts and parallel execution. The user experience is also accounted for, with responsive web UI and source code analysis to automatically find out the number and type of input streams. Finally, we test the system by utilizing the library to parse and pre-process data and implement an inference algorithm to run on the pre-processed data. A demo on large-scale soil moisture data is also provided.
