Sakhanda Wire
NVDA $154.51 -1.04% MSFT $506.49 -2.22% GOOGL $281.27 -0.86% META $621.51 -0.78% AMZN $256.90 -0.33%
← Back to the news

Meta FAIR releases e-means clustering algorithm in FAISS 1.16.0

Meta FAIR researchers have released e-means, a mini-batch replacement for Lloyd’s k-means algorithm. The method is available in FAISS 1.16.0 under faiss.contrib.e_means and supports CPU or GPU execution with PyTorch tensors or NumPy arrays. It is intended for clustering tasks used in vector indexes, quantizers and data-curation pipelines.

E-means updates each cluster’s accumulated sum and count with exponential moving averages, allowing older assignments to lose influence. Its batch size and initial step are selected through scaling laws based on dataset size, cluster count and the planned number of epochs, with no additional tuning parameters reported. The step follows a cosine schedule, and small clusters can be revived by splitting larger ones.

In the authors’ tests, e-means reached the loss of Lloyd’s 25-epoch run in 1.4 to 2.6 epochs across 24 configurations. The reported speedups were 10-45 times on GPUs compared with Flash-KMeans and 8-15 times on CPUs compared with Faiss k-means. Tests ranged from 100K to 10 billion points and up to 16.7M clusters. Runs with 16M centroids require 64 GiB of float32 accumulators and sharding across GPUs.

Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.

Originally published by MarkTechPost on

Read the original on MarkTechPost ↗

The reporting and the image belong to MarkTechPost and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news