TL;DR of the paper, generated automatically.

Meta's e-means clustering algorithm reaches lower loss than k-means, in far fewer epochs.

• Speed: beats Lloyd's 25-epoch loss in 1.4 to 2.6 epochs.
• GPU time: 10-45x faster than state-of-the-art k-means implementations.
• Scale: 10 billion embeddings into 16 million clusters in 2.5 epochs.