Breaking the Reclustering Barrier in Centroid-based Deep Clustering

2024-11-04

arXiv.org

AI Summary

A prevalent “re-clustering barrier” in centroid-based deep clustering—characterized by early performance saturation and failure of periodic centroid re-initialization—is systematically identified for the first time as stemming from premature convergence to and rigid dependence on initial cluster centroids. To address this, we propose BRB (Bias-Resistant Balancing), a novel algorithm that decouples centroid update from feature learning, introduces a dynamic, adaptive centroid re-initialization mechanism, and integrates contrastive learning loss to enhance semantic discriminability. BRB enables end-to-end training from scratch without pretraining or external supervision. Evaluated on multiple benchmark datasets (e.g., MNIST, CIFAR-10, STL-10), BRB consistently improves clustering accuracy, effectively overcoming the performance bottleneck and achieving state-of-the-art (SOTA) results.

Technology Category

• Machine Learning: Clustering
• Computer Vision: Representation Learning for Vision
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

This work investigates an important phenomenon in centroid-based deep clustering (DC) algorithms: Performance quickly saturates after a period of rapid early gains. Practitioners commonly address early saturation with periodic reclustering, which we demonstrate to be insufficient to address performance plateaus. We call this phenomenon the"reclustering barrier"and empirically show when the reclustering barrier occurs, what its underlying mechanisms are, and how it is possible to Break the Reclustering Barrier with our algorithm BRB. BRB avoids early over-commitment to initial clusterings and enables continuous adaptation to reinitialized clustering targets while remaining conceptually simple. Applying our algorithm to widely-used centroid-based DC algorithms, we show that (1) BRB consistently improves performance across a wide range of clustering benchmarks, (2) BRB enables training from scratch, and (3) BRB performs competitively against state-of-the-art DC algorithms when combined with a contrastive loss. We release our code and pre-trained models at https://github.com/Probabilistic-and-Interactive-ML/breaking-the-reclustering-barrier .

Problem

• Addresses performance saturation in centroid-based deep clustering.
• Identifies and overcomes the 'reclustering barrier' phenomenon.
• Introduces BRB algorithm for continuous clustering adaptation.

Innovation

• Avoids early over-commitment to initial clusterings
• Enables continuous adaptation to reinitialized targets
• Combines with contrastive loss for competitive performance