Inference techniques in low-cost sensor networksInference techniques in low-cost sensor networksΤεχνικές συμπερασμού σε δίκτυα αισθητήρων χαμηλού κόστους Μεταπτυχιακή Διατριβή Master Thesis 2023-04-052023enDistributed execution of algorithms across resource-constrained terminals has become increasingly popular, especially when fault tolerance is required. Asynchronous operation is brought to light in such scenarios, and in particular, probabilistic asynchronous operation, which models the failure probability of each terminal. The focus of this work is on the affine update model, which is applicable to a wide range of distributed inference algorithms. Applications include estimation of the average, solving linear systems, linear minimum mean square error estimation and spectral clustering, presented in detail in this thesis. Multimodal inference is also investigated, where two or more alternate sources of data are exploited for increased prediction accuracy. In that context, two variations of linear regression are presented, with uniform or Gaussian prior, which are equivalent to iterative affine updates. Furthermore, this work offers an asymptotic analysis for the arithmetic mean of the state vector, across a finite number of experiments, for the discovery of fixed points. It is shown that there are cases where the arithmetic mean behaves differently than the expected mean, and a sufficient condition is provided for convergence of the arithmetic mean to a fixed point. The lack of necessity for this condition is explained and subcases where the arithmetic mean converges, diverges, or has unpredictable behaviour are highlighted. Additionally, cases where the individual iterations never converge but their arithmetic mean does and offers fixed point are offered. Simulations corroborate the theoretical findings for various affine model setups. Finally, implementation details of a distributed low-cost sensor network are presented; the low-cost sensor network estimates the arithmetic mean of temperature across the network, by executing average consensus. The implementation is divided into hardware, network and software layers, and each layer is presented separately.http://creativecommons.org/licenses/by/4.0/Πολυτεχνείο Κρήτης::Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών ΥπολογιστώνApostolakis_Georgios_MSc_2023.pdfChania [Greece]Library of TUC2023-04-05application/pdf2.5 MBembargo Apostolakis Georgios Αποστολακης Γεωργιος Bletsas Aggelos Μπλετσας Αγγελος Lagoudakis Michail Λαγουδακης Μιχαηλ Deligiannakis Antonios Δεληγιαννακης Αντωνιος Πολυτεχνείο Κρήτης Technical University of Crete Συμπερασμός Inference Δίκτυα αισθητήρων Sensor networks Αφινικό μοντέλο Affine model Asynchrony Convergence region Asymptotic properties Arithmetic mean Distributed implementation περιοχή σύγκλισης αριθμητικός μέσος όρος κατανεμημένη υλοποίηση