Class Centroid

Class centroids, representing the central tendency of feature vectors within a data class, are crucial for various machine learning tasks. Current research focuses on improving centroid estimation robustness, particularly in the face of noisy labels, long-tailed data distributions, and cross-domain discrepancies, often employing techniques like weighted averaging, K-means clustering, and adaptive quantization. These advancements enhance model performance in applications such as image classification, large language model optimization, and unsupervised person re-identification, ultimately leading to more accurate and reliable machine learning systems.