Spectral Bridges

TL;DR

Spectral Bridges is a new clustering algorithm that combines spectral and Voronoi-based methods, offering a fast, robust, and versatile approach capable of handling complex, non-convex data structures in large-scale datasets.

Contribution

It introduces a non-parametric clustering method inspired by SVM margins, integrating Voronoi subdivision and spectral techniques with minimal hyperparameters.

Findings

Effective on large-scale real-world datasets

Handles intricate non-convex cluster structures

Demonstrates robustness and speed in diverse scenarios

Abstract

In this paper, Spectral Bridges, a novel clustering algorithm, is introduced. This algorithm builds upon the traditional k-means and spectral clustering frameworks by subdividing data into small Vorono\"i regions, which are subsequently merged according to a connectivity measure. Drawing inspiration from Support Vector Machine's margin concept, a non-parametric clustering approach is proposed, building an affinity margin between each pair of Vorono\"i regions. This approach is characterized by minimal hyperparameters and delineation of intricate, non-convex cluster structures. The numerical experiments underscore Spectral Bridges as a fast, robust, and versatile tool for sophisticated clustering tasks spanning diverse domains. Its efficacy extends to large-scale scenarios encompassing both real-world and synthetic datasets. The Spectral Bridge algorithm is implemented both in Python (<https://pypi.org/project/spectral-bridges>) and R <https://github.com/cambroise/spectral-bridges-Rpackage>).