E-means: 10x faster k-means at equal loss

E-means is a new k-means clustering algorithm that achieves 10-45× faster computation on GPUs and 8-15× faster on CPUs compared to state-of-the-art implementations, while simultaneously achieving better or equal clustering loss. The method replaces the traditional Lloyd algorithm with mini-batch updates using exponential moving averages, automatic hyperparameter tuning via scaling laws, and improved initialization and finalization strategies, eliminating the need for manual hyperparameter tuning.