Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Adaptive Resonance Theory-based Topological Clustering Algorithm with a Self-Adjusting Vigilance Parameter

Naoki Masuyama, Yuichiro Toda, Yusuke Nojima, Hisao Ishibuchi

TL;DR

IDAT addresses streaming clustering under evolving data by embedding an ART-based topological clustering framework with parameter-free adaptation. It automatically tunes its recalculation interval $\Lambda$ and vigilance threshold $V_{\text{threshold}}$ through a diversity-driven mechanism, updating topology in a single-pass online fashion. Across 24 real-world datasets, IDAT achieves superior clustering quality and continual-learning performance, while maintaining a compact topology and mitigating catastrophic forgetting. The approach offers practical benefits for open-world, many-class scenarios by removing manual hyperparameter tuning and enabling robust online clustering in changing environments.

Abstract

Clustering in stationary and nonstationary settings, where data distributions remain static or evolve over time, requires models that can adapt to distributional shifts while preserving previously learned cluster structures. This paper proposes an Adaptive Resonance Theory (ART)-based topological clustering algorithm that autonomously adjusts its recalculation interval and vigilance threshold through a diversity-driven adaptation mechanism. This mechanism enables hyperparameter-free learning that maintains cluster stability and continuity in dynamic environments. Experiments on 24 real-world datasets demonstrate that the proposed algorithm outperforms state-of-the-art methods in both clustering performance and continual learning capability. These results highlight the effectiveness of the proposed parameter adaptation in mitigating catastrophic forgetting and maintaining consistent clustering in evolving data streams. Source code is available at https://github.com/Masuyama-lab/IDAT

An Adaptive Resonance Theory-based Topological Clustering Algorithm with a Self-Adjusting Vigilance Parameter

TL;DR

IDAT addresses streaming clustering under evolving data by embedding an ART-based topological clustering framework with parameter-free adaptation. It automatically tunes its recalculation interval and vigilance threshold through a diversity-driven mechanism, updating topology in a single-pass online fashion. Across 24 real-world datasets, IDAT achieves superior clustering quality and continual-learning performance, while maintaining a compact topology and mitigating catastrophic forgetting. The approach offers practical benefits for open-world, many-class scenarios by removing manual hyperparameter tuning and enabling robust online clustering in changing environments.

Abstract

Clustering in stationary and nonstationary settings, where data distributions remain static or evolve over time, requires models that can adapt to distributional shifts while preserving previously learned cluster structures. This paper proposes an Adaptive Resonance Theory (ART)-based topological clustering algorithm that autonomously adjusts its recalculation interval and vigilance threshold through a diversity-driven adaptation mechanism. This mechanism enables hyperparameter-free learning that maintains cluster stability and continuity in dynamic environments. Experiments on 24 real-world datasets demonstrate that the proposed algorithm outperforms state-of-the-art methods in both clustering performance and continual learning capability. These results highlight the effectiveness of the proposed parameter adaptation in mitigating catastrophic forgetting and maintaining consistent clustering in evolving data streams. Source code is available at https://github.com/Masuyama-lab/IDAT

Paper Structure

This paper contains 31 sections, 24 equations, 11 figures, 23 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Batch Clustering
  4. Growing Self-Organizing Clustering
  5. Deep Learning-based Clustering
  6. Proposed Algorithm
  7. Overview
  8. Initialization
  9. Clustering Step
  10. Diversity-Driven Adaptation of $\Lambda$ and $V_{\text{threshold}}$
  11. Diversity Assessment via Matrix Stability
  12. Adaptive Adjustment of $\Lambda$
  13. Estimating $V_{\text{threshold}}$ from Similarity Matrix Stability
  14. Prediction Procedure
  15. Computational Complexity
  16. ...and 16 more sections

Figures (11)

  • Figure 1: Critical difference diagrams based on the average ARI and AMI in the stationary setting.
  • Figure 2: Critical difference diagrams based on the average ARI and AMI in the nonstationary setting.
  • Figure 3: Critical difference diagrams based on AI-ARI, AI-AMI, BWT-ARI, and BWT-AMI in the nonstationary setting.
  • Figure 4: Critical difference diagrams based on ARI, AMI, and continual performance metrics in the nonstationary setting for the ablation study. The notation (500) indicates $\Lambda_{\text{init}} = 500$, and its absence implies $\Lambda_{\text{init}} = 2$.
  • Figure 5: Histories of $\Lambda$ and $V_{\text{threshold}}$ for STL10 in the nonstationary setting. In (g) and (h), the two lines are constant and would overlap. Thus, an offset is applied to improve visibility.
  • ...and 6 more figures