Quantization by Clustering

K-means Quantization

Clustering algorithm applied to neural network weight quantization by partitioning the weight space into K clusters. Weights are then represented by their respective cluster centroids, thus reducing the required precision.

← Zurück