We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

cs

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo

Computer Science > Machine Learning

Title: Efficient Graph Field Integrators Meet Point Clouds

Abstract: We present two new classes of algorithms for efficient field integration on graphs encoding point clouds. The first class, SeparatorFactorization(SF), leverages the bounded genus of point cloud mesh graphs, while the second class, RFDiffusion(RFD), uses popular epsilon-nearest-neighbor graph representations for point clouds. Both can be viewed as providing the functionality of Fast Multipole Methods (FMMs), which have had a tremendous impact on efficient integration, but for non-Euclidean spaces. We focus on geometries induced by distributions of walk lengths between points (e.g., shortest-path distance). We provide an extensive theoretical analysis of our algorithms, obtaining new results in structural graph theory as a byproduct. We also perform exhaustive empirical evaluation, including on-surface interpolation for rigid and deformable objects (particularly for mesh-dynamics modeling), Wasserstein distance computations for point clouds, and the Gromov-Wasserstein variant.
Subjects: Machine Learning (cs.LG)
Journal reference: ICML 2023
Cite as: arXiv:2302.00942 [cs.LG]
  (or arXiv:2302.00942v6 [cs.LG] for this version)

Submission history

From: Han Lin [view email]
[v1] Thu, 2 Feb 2023 08:33:36 GMT (9928kb,D)
[v2] Sun, 5 Feb 2023 20:12:24 GMT (9923kb,D)
[v3] Wed, 12 Apr 2023 22:27:17 GMT (10017kb,D)
[v4] Sat, 10 Jun 2023 01:29:45 GMT (10017kb,D)
[v5] Wed, 21 Jun 2023 02:34:32 GMT (5018kb,D)
[v6] Wed, 4 Oct 2023 19:17:43 GMT (4990kb,D)

Link back to: arXiv, form interface, contact.