BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Computational Imaging: Algorithms & Applications
DTSTART:20190821T090000
DTEND:20190821T110000
DTSTAMP:20260408T134507Z
UID:6a4d19dce8d0159516f3d5b3efc9fa6fa1ada05192dcc380b49a05f1
CATEGORIES:Conferences - Seminars
DESCRIPTION:Sepand Kashani-Akhavan\nEDIC candidacy exam\nExam president: P
 rof. Pierre Vandergheynst\nThesis advisor: Prof. Martin Vetterli\nThesis c
 o-advisor: Prof. Paul Hurley\nCo-examiner: Prof. Jean-Philippe Thiran\n\nA
 bstract\nSignal and image analysis are key investigative techniques for ex
 tracting knowledge of natural\nphenomena in various scientific fields.  O
 ften however the processes of interest cannot be directly\nobserved.  Ins
 tead\, indirect evidence of the phenomena responsible for image formation 
 can be\nmeasured and used to estimate the original image.\n\nAdvances in i
 maging science owe a large part to breakthroughs in instrumentation.  Nev
 ertheless the\ntraditional imaging landscape is challenged from all direct
 ions today:  on the one hand applications\nsuch as microscopy and astrono
 my constantly seek to resolve features beyond resolution limits of\ntheir 
 instruments.  On the other hand medical imaging often has limited measure
 ment time for\npractical purposes and must learn to do more with less.  F
 inally\, conventional reconstruction\nalgorithms that work with current ge
 neration instruments may no longer be tractable at larger\nscales.  To br
 idge the gaps above\, the trend is to involve computation in the imaging c
 hain\, by\ncombining knowledge of the acquisition system with mathematical
  models and optimization theory to\nfuel the next revolution in imaging sc
 ience.\n\nThis paper analyses three background works on classical and lear
 ning-based inverse problems in\ncomputational imaging.  We first look at 
 recent work by members of the Event Horizon Telescope (EHT)\nto image blac
 k holes through radio-interferometry using classical methods.  Moving ont
 o\nlearning-based methods\, we look at design criteria to extend Convoluti
 onal Neural Networks (CNNs) to\nnon-Euclidean domains\, specifically to di
 scretized spherical manifolds.  Finally\, we revisit\nclassical methods w
 ith a focus on non-standard algorithmic solvers for compressed sensing pro
 blems.\n\nBackground papers\n1. Super-resolution Interferometric Imaging w
 ith Sparse Modeling using\nTotal Squared Variation -- Application to Imagi
 ng the Black Hole Shadow\n[Kuramochi\, Akiyama\, Ikeda\, Tazaki\, Fish\, P
 u\, Asada\, Honma].\n2. DeepSphere: Efficient spherical Convolutional Neur
 al Network with\nHEALpix sampling for cosmological applications [Perraudin
 \, Defferrard\,\nKacprzak\, Sgier].\n3. Message-passing algorithms for com
 pressed sensing [Donoho\, Maleki\,\nMontanari].
LOCATION:BC 010 https://plan.epfl.ch/?room==BC%20010
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR