BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//TUC//Events//EN
CALSCALE:GREGORIAN
BEGIN:VTIMEZONE
TZID:Europe/Athens
TZNAME:EEST
DTSTART:19700329T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0300
TZNAME:EET
DTSTART:19701025T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CREATED:20220923T083712Z
LAST-MODIFIED:20220923T083712Z
DTSTAMP:20260415T042007Z
UID:1776216007@tuc.gr
SUMMARY:Παρουσίαση Διπλωματικής Εργασίας κ. 
 Καριωτάκη Εμμανουήλ - Σχολή ΗΜΜΥ
LOCATION:
DESCRIPTION:https://www.tuc.gr/el/to-polytechnei
 o/ilektronikes-ypiresies/imerologio/
 imerologio-ekdiloseon-1?tx_tucevents
 2_tuceventsdisplay%5Baction%5D=show&
 tx_tucevents2_tuceventsdisplay%5Bcon
 troller%5D=Event&tx_tucevents2_tucev
 entsdisplay%5Bevent%5D=5795&cHash=04
 d4e298864411db0004706daa7b13d7\nΠΟΛΥ
 ΤΕΧΝΕΙΟ ΚΡΗΤΗΣ\n Σχολή Ηλεκτρολόγων 
 Μηχανικών και Μηχανικών Υπολογιστών\
 n Πρόγραμμα Προπτυχιακών Σπουδών\n Π
 ΑΡΟΥΣΙΑΣΗ ΔΙΠΛΩΜΑΤΙΚΗΣ ΕΡΓΑΣΙΑΣ\n ΕΜ
 ΜΑΝΟΥΗΛ ΚΑΡΙΩΤΑΚΗ\n με θέμα\n Ομαδοπ
 οίηση Αλγορίθμων Συμπερασμού σε Δίκτ
 υα Επικοινωνιών \n Clustering of Inf
 erence Algorithms in Communication N
 etworks\n Εξεταστική Επιτροπή\n Καθη
 γητής Άγγελος Μπλέτσας (επιβλέπων)\n
  Καθηγητής Μιχαήλ Ζερβάκης\n Καθηγητ
 ής Γεώργιος Καρυστινός\n Abstract\n 
 This work offers an algorithmic fram
 ework for in-network inference, usin
 g message passing among ambiently po
 wered wireless sensor network (WSN) 
 terminals. The stochastic nature of 
 ambient energy harvesting dictates i
 ntermittent operation of each WSN te
 rminal and as such, the message pass
 ing inference algorithms should be r
 obust to asynchronous operation. A v
 ersion of Gaussian Belief Algorithm 
 (GBP) is described, which can be red
 uced to an affine fixed point (AFP) 
 problem, used to solve linear system
 s of equations. To achieve this, we 
 have to cluster the Probabilistic Gr
 aphical Model (PGM) behind GBP, in o
 rder to map it to the WSN terminals.
  We propose two different clustering
  approaches, namely edge and node cl
 ustering. For the first approach, we
  explain the reasons why a previous 
 method does not produce the expected
  results and we offer another method
 , which performs better. We also exp
 lain limitations of edge-based clust
 ering. On the other hand, node clust
 ering has a clear metric for perform
 ance, which is relevant to the numbe
 r of edges connecting the different 
 clusters. For this approach, we util
 ize three different clustering algor
 ithms, the k-means, the spectral clu
 stering and an autonomous, in-networ
 k clustering algorithm. Furthermore,
  we show in both theory and simulati
 on that there is strong connection b
 etween spectral radius and the conve
 rgence rate of AFP problems with pro
 babilistic asynchronous scheduling. 
 The latter corroborates known theory
  for synchronous scheduling. Interes
 tingly, it is shown through simulati
 ons that different clustering offers
  similar convergence rate, when prob
 abilistic asynchronous scheduling is
  utilized with carefully selected pr
 obabilities that accelerate converge
 nce rate in the mean sense. Finally,
  we show an existing distinction bet
 ween convergence rate and energy con
 sumption of the network and we prese
 nt experimental results comparing th
 e different clustering methods. In m
 ost cases, spectral clustering outpe
 rforms the rest, with reduced energy
  consumption (by a factor of 2 compa
 red to k-means in specific cases).\n
  Meeting ID: 947 4583 0692\n Passwor
 d: 034448\n
STATUS:CONFIRMED
ORGANIZER;RSVP=FALSE;CN=TUC;CUTYPE=TUC:mailto:webmaster@tuc.gr
DTSTART:20220926T160000
DTEND:20220926T170000
TRANSP:OPAQUE
CLASS:DEFAULT
END:VEVENT
END:VCALENDAR