Optimizing neighbor connections in nearest neighbor graphs

Paper Title:

Optimizing $k$ in $k$NN Graphs with Graph Learning Perspective

Published on:

16 January 2024

Primary Category:

Machine Learning

Paper Authors:

Asuka Tamaru,

Junya Hara,

Hiroshi Higashi,

Yuichi Tanaka,

Antonio Ortega

Bullets

Key Details

•

Proposes method to optimize k differently for each node in kNN graphs

•

Formulates discrete optimization to determine best k based on sum of distances

•

Reveals relationship between proposed method and graph learning approaches

•

Produces sparse graphs that connect more similar nodes

•

Improves performance for tasks like point cloud denoising

Explore the topics in this paper

graph learning

k-nearest neighbor graphs

point cloud denoising

similarity graphs

sparse graphs

AI generated summary

Optimizing neighbor connections in nearest neighbor graphs

This paper proposes a method to optimize the number of neighbor connections (k) for each node in k-nearest neighbor graphs. It formulates an optimization problem to determine the best k for each node based on sum of distances constraints. The method connects more similar nodes while keeping graphs sparse. Experiments show it improves graph learning and point cloud denoising applications.