CLUBench: A Clustering Benchmark

2026-05-28 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, quick

Summary

CLUBench is a new, comprehensive clustering benchmark designed to address the lack of systematic, large-scale empirical evaluation across diverse clustering algorithms. It evaluates 24 algorithms, including conventional, deep learning, and foundation model-based methods, on 131 datasets spanning tabular, text, and image data, involving 178,815 experiments. Key analyses reveal that deep clustering methods do not significantly outperform top conventional algorithms like KMeans and SpeClu on average. For image and text tasks, combining pretrained embeddings with conventional algorithms proves effective and efficient. The benchmark also highlights that clustering remains a challenging problem, even with advanced foundation models, and proposes using low-rank structures in performance matrices for efficient evaluation and model selection.

Key takeaway

For Machine Learning Engineers evaluating clustering solutions, you should prioritize conventional algorithms like KMeans or SpeClu, especially when combined with pretrained embeddings for image and text data, as deep clustering methods show no significant average performance advantage. Utilize CLUBench's findings to efficiently approximate overall performance and guide model selection by analyzing low-rank structures in performance matrices, optimizing your resource allocation.

Key insights

CLUBench provides a comprehensive benchmark revealing conventional clustering often matches deep learning, especially with pretrained embeddings.

Principles

• Deep clustering lacks average advantage.
• Pretrained embeddings boost conventional methods.
• Clustering remains a hard problem.

Method

The benchmark proposes using low-rank structures in cross-model performance matrices to efficiently approximate overall performance evaluation and enable model selection across hyperparameter configurations.

In practice

• Combine pretrained embeddings with KMeans.
• Consider SpeClu for image/text data.
• Utilize performance matrix low-rank structures.

Topics

• Clustering Benchmarks
• Deep Clustering
• Foundation Models
• Pretrained Embeddings
• KMeans
• SpeClu
• Performance Evaluation

Best for: Research Scientist, AI Engineer, NLP Engineer, AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• Structure Retention in Embedding Spaces as a Predictor of Benchmark Performance — Embedding space organization, specifically nearest-neighbor overlap and ICA magnitude, strongly predicts model performance.
• Scalable Model-Based Clustering with Sequential Monte Carlo — A novel SMC algorithm improves scalability for online clustering by decomposing problems into independent subproblems.
• Lumbermark: Resistant Clustering by Chopping Up Mutual Reachability Minimum Spanning Trees — Lumbermark is a robust, divisive clustering algorithm that leverages mutual reachability MSTs to detect varied clusters with user-specified counts.
• Top 20 Unsupervised Learning Interview Questions and Answers (Part 1 of 2) — Unsupervised learning extracts structure from unlabeled data using statistical patterns for various ML applications.
• A Multi-Dataset Benchmark of Multiple Instance Learning for 3D Neuroimage Classification — Simple mean pooling MIL offers efficient and competitive 3D neuroimage classification, often outperforming complex alternatives.
• Smart Data Grouping: Organizing Networks Without Guesswork — A new differentiable structural information formulation enhances deep graph clustering via LSEnet.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.