BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Geometric deep learning on the sphere: scalable and equivariant sp
 herical CNNs - Jason McEwen
DTSTART:20220923T090000Z
DTEND:20220923T100000Z
UID:TALK180308@talks.cam.ac.uk
CONTACT:Will Handley
DESCRIPTION:Many problems across computer vision and the natural sciences 
 require the analysis of spherical data. For example\, the cosmic microwave
  background (CMB) relic radiation from the Big Bang is acquired on the sph
 ere\, as are 360° photos and videos that are prevalent in virtual reality
 . To leverage the potential of deep learning in these areas and others\, g
 eometric deep learning techniques defined natively on the sphere are requi
 red. In this talk I will discuss connections between physics and deep lear
 ning\, and in particular the role of symmetry in representation learning. 
 I will highlight the importance of encoding the symmetries and geometric p
 roperties of the sphere in spherical deep learning constructions\, in orde
 r to capture rotational equivariance so that representations may be learne
 d effectively. I will discuss various frameworks for constructing spherica
 l convolutional neural networks (CNNs)\, including continuous\, discrete a
 nd hybrid approaches\, and their pros and cons. I will concisely review ou
 r recent works on Efficient Generalized Spherical CNNs ( https://arxiv.org
 /abs/2010.11661)\, Scattering Networks on the Sphere ( https://arxiv.org/a
 bs/2102.02828)\, and Scalable and Equivariant Spherical CNNs (yet to be pu
 blished). I will present the application of our frameworks to numerous sph
 erical deep learning benchmark tasks\, on all of which we achieve the stat
 e-of-the-art performance.\n
LOCATION:Large Martin Ryle Seminar Room\, KICC
END:VEVENT
END:VCALENDAR