BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Inverse consistency and global convergence of ResNets - Francois-X
 avier Vialard (Université Gustave Eiffel)
DTSTART:20220601T130000Z
DTEND:20220601T140000Z
UID:TALK173954@talks.cam.ac.uk
CONTACT:Willem Diepeveen
DESCRIPTION:In this talk\, I will discuss two very different applications 
 related by the simple idea of invertible transformations.\n\n- The first t
 opic is an analysis of inverse consistency penalty in image matching in co
 njunction with the use of neural networks. We show that neural networks fa
 vours the  emergence of smooth transformation for the inverse consistency.
  Experimentally\, we show that this behaviour is fairly stable with respec
 t to the chosen architecture. This is joint work with H. Greer\, R. Kwitt 
 and M. Niethammer.\n\n-The second topic is an analysis of global convergen
 ce of residual networks when the residual block is parametrized via reprod
 ucing kernel Hilbert space vector field. We prove that the resulting probl
 em satisfies the so-called Polyak-Lojasiewicz property\, for instance ensu
 ring global convergence if the iterates are bounded. We show that this pro
 perty applies in a continuous limit as well as in the fully discrete setti
 ng. This is joint work with R. Barboni and G. Peyré.\n\n\nJoin Zoom Meeti
 ng\nhttps://maths-cam-ac-uk.zoom.us/j/94812219444?pwd=K00vZUVUU2NDbHozR2h1
 UzdLRlI1QT09\n\nMeeting ID: 948 1221 9444\nPasscode: 485548
LOCATION:Virtual
END:VEVENT
END:VCALENDAR