BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Apprentissage sur des graphes attribués
DTSTART:20230622T133000
DTEND:20230622T153000
DTSTAMP:20260610T124957Z
UID:b6d70922c983babb1eb927c6bbba829f42d7607e75e092c6bb023a0f
CATEGORIES:Conferences - Seminars
DESCRIPTION:Odilon Duranthon\nEDIC candidacy exam\nExam president: Prof. N
 icolas Macris\nThesis advisor: Prof. Lenka Zdeborová\nCo-examiner: Prof. 
 Pascal Frossard\n\nAbstract\nIn the frame of inference on graphs growing i
 nterest has been given to synthetic datasets generated by tractable models
 . For such models the average performance of simple graph neural networks 
 (GNNs) can be predicted and an optimal algorithm based on belief-propagati
 on (BP) can be derived. We derive the optimal solution for the most common
 ly used model and we show that state-of-the-art GNNs are far from the opti
 mality. We propose to build an architecture inspired by BP and we hope it 
 will close the gap.\n\nBackground papers\n1. « Contextual Stochastic Bloc
 k Models » https://dspace.mit.edu/bitstream/handle/1721.1/138073/8077-con
 textual-stochastic-block-models.pdf\n2. « Statistical Mechanics of Genera
 lization In Graph Convolution Networks »https://arxiv.org/abs/2212.13069\
 n3. « AMP-Inspired Deep Networks for Sparse Linear Inverse Problems » ht
 tps://arxiv.org/abs/1612.01183\n\n 
LOCATION:
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR