Anchor point selection is the process of identifying or generating a compact set of representative reference points from a dataset to approximate its underlying distribution and manifold structure. In large-scale and graph-based machine learning methods, such as anchor-based clustering and dimensionality reduction, this technique reduces the computational and memory burdens of calculating pairwise relationships across all data samples. By measuring similarities primarily between data points and the selected anchors, algorithms can lower their complexity from quadratic to linear relative to the total number of samples. Selection strategies range from heuristic methods like random sampling and clustering centroids to continuous, adaptive optimization integrated directly into graph learning frameworks.
