Non-iterative Label Propagation in Optimal Leading Forest

TL;DR

This paper introduces LaPOLeaF, a non-iterative label propagation algorithm based on an optimal leading forest structure, significantly improving efficiency and scalability in graph-based semi-supervised learning.

Contribution

It proposes a novel non-iterative label propagation method using the optimal leading forest, addressing efficiency and scalability issues in traditional GSSL.

Findings

Achieves high efficiency in label propagation

Scales effectively to large datasets

Demonstrates promising results on big data experiments

Abstract

Graph based semi-supervised learning (GSSL) has intuitive representation and can be improved by exploiting the matrix calculation. However, it has to perform iterative optimization to achieve a preset objective, which usually leads to low efficiency. Another inconvenience lying in GSSL is that when new data come, the graph construction and the optimization have to be conducted all over again. We propose a sound assumption, arguing that: the neighboring data points are not in peer-to-peer relation, but in a partial-ordered relation induced by the local density and distance between the data; and the label of a center can be regarded as the contribution of its followers. Starting from the assumption, we develop a highly efficient non-iterative label propagation algorithm based on a novel data structure named as optimal leading forest (LaPOLeaF). The major weaknesses of the traditional GSSL are addressed by this study. We further scale LaPOLeaF to accommodate big data by utilizing block distance matrix technique, parallel computing, and Locality-Sensitive Hashing (LSH). Experiments on large datasets have shown the promising results of the proposed methods.