SuperEMeans makes e-means clustering 3× faster with no quality loss
SuperEMeans: Making e-means 3× faster
e-means promised Lloyd's-quality clustering in just 2 epochs, and SuperKMeans sped up k-means by skipping vector dimensions. Combining both ideas, SuperEMeans clusters 5M OpenAI and 10M Cohere embeddings up to 3× faster than e-means at 5 epochs, with identical recall@100. The author also finds e-means' revival threshold too conservative, leaving bloated clusters that inflate scan costs in vector search.
e-means reaches a lower WCSS (MSE), since that objective doesn't reward balanced clusters, while vector search pays for every point it scans.