Attendees at a technical conference focus attentively on a detailed presentation titled "Gromov-Wasserstein for Encoding Structural Priors." The lecture explores a general formulation for discrete Gromov-Wasserstein (GW) problems, delving into complex mathematical equations involving cost matrices and loss functions. Captured from the audience's perspective, several individuals are seen taking photos of the presentation slides with their smartphones, ensuring they capture the intricate details displayed on the large screens. The venue is a spacious hall with high ceilings and industrial-style lighting, hosting an engaged and curious audience eager to learn from the insightful session. Furthermore, the backdrop proudly displays "CVPR Seattle," indicating the conference's focus on computer vision and pattern recognition, held in Seattle.
Text transcribed from the image:
Gromov-Wasserstein for
Encoding Structural Priors
A relatively) general formulation for (discrete) GW problems:
minimize
c
CVPR
JUNE 17-21, 2024
SEATTLE, W
Gromov-Wasserstein for
Encoding Structural Priors
A relatively) p
screte) GW problems
St
TI-I
Cost matrices Cea and c c R
"Loss function between cost matry
Cost m
- "Loss"
ments La Rx-
Gromov-Wasserstein for
Encoding Structural Priors
A (relatively) general formulation for (discrete) GW problems:
minimize Σi,ke[N] L(CC)TijTkl
TERNXK
s.t.
j.LE KI
T1K = 1N,
T1N=1K,
> Cost matrices C E IRNXN and Ca E RKXK
Nation an
University
> "Loss" function between cost matrix elements L: IR XR-R
Ming Xu and Stephen Gould. Temporally Consistent Unbalanced Optimal Transport for Unsupervised
SUPP