Meta FAIR's e-means beats k-means in speed and quality

Meta FAIR's e-means uses exponential moving averages to outperform classic k-means in both quality and speed, training in 1.4-2.6 epochs versus 25 and running 10-45x faster on GPU. An IVF index-accelerated trick further cuts compute by 4.3x on billion-vector datasets.

2026-11-10 ~ 2026-11-10 · 2 related posts

• e-means paper trick: IVF approximate assignment cuts clustering cost 34x at 10B vectors — antoine_chaffin · 2026-11-10
• Meta FAIR's e-means beats k-means in 1/10 the epochs with 10-45x GPU speedups — kastnerkyle · 2026-11-10