AdaGraph: Graph-Native Clustering Algorithm Overcomes Curse of Dimensionality

AdaGraph is a graph-native clustering algorithm introduced in arXiv:2605.16320, developed under the Structure-Centric Machine Learning (SC-ML) paradigm. Unlike traditional geometry-centric methods, AdaGraph operates entirely within the kNN graph topology, preserving relational structure in high-dimensional spaces where Euclidean distance fails. It requires no pre-specified number of clusters, handles noise natively, and scales via the SLCD framework. Its unsupervised tuning uses Graph-SCOPE, a topology-based cluster validity index. On 10 synthetic benchmarks from d=10 to d=5000, Graph-SCOPE achieved superior performance.