SuperEMeans: Making e-means 3× faster
Summary
This blog post by Leonardo Kuffo introduces SuperEMeans, an open-source approach that fuses SuperKMeans with e-means to achieve up to 3× faster clustering of high-dimensional vector embeddings without sacrificing quality. It benchmarks on OpenAI and MS MARCO datasets, compares against Lloyd’s k-means and standard e-means, and discusses revival thresholds to improve cluster balance. The post also provides reproducibility resources and links to related open-source work.