BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Artificial Intelligence Research Group Talks (Comp
uter Laboratory)
SUMMARY:Approximate Equivariance SO(3) Needlet Convolution
- Kai Yi\, University of New South Wales (UNSW) i
n Sydney
DTSTART;TZID=Europe/London:20221130T170000
DTEND;TZID=Europe/London:20221130T180000
UID:TALK193115AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/193115
DESCRIPTION:This paper develops a rotation-invariant needlet c
onvolution for rotation group SO(3) to distill mul
tiscale information of spherical signals.\nThe sph
erical needlet transform is generalized from $\\sS
^2$ onto the SO(3) group\, which decomposes a sphe
rical signal to approximate and detailed spectral
coefficients by a set of tight framelet operators.
The spherical signal during the decomposition and
reconstruction achieves rotation invariance. \nBa
sed on needlet transforms\, we form a Needlet appr
oximate Equivariance Spherical CNN (NES) with mult
iple SO(3) needlet convolutional layers. The netwo
rk establishes a powerful tool to extract geometri
c-invariant features of spherical signals. \nThe m
odel allows sufficient network scalability with mu
lti-resolution representation. A robust signal emb
edding is learned with wavelet shrinkage activatio
n function\, which filters out redundant high-pass
representation while maintaining approximate rota
tion invariance. \nThe NES achieves state-of-the-a
rt performance for quantum chemistry regression an
d Cosmic Microwave Background (CMB) delensing reco
nstruction\, which shows great potential for solvi
ng scientific challenges with high-resolution and
multi-scale spherical signal representation.\n\n \
n\n*Bio:*\n\nKai Yi is a third-year Ph.D. student
majoring in mathematics at the School of Mathemati
cs and Statistics in the University of New South W
ales (UNSW) in Sydney\, Australia. His research in
terests lie in geometric deep learning and Bayesia
n statistics. He has applied geometric deep learni
ng methods to estimating gravitational lensing par
ameters in cosmology and used VAE for inpainting C
MB maps.
LOCATION:Lecture Theatre 2
CONTACT:Pietro Lio
END:VEVENT
END:VCALENDAR