顶刊TPAMI 2026！GBSK：面向复杂分布的大规模高效聚类
Ranking
No observed public metrics; popularity remains neutral/archived.
Merged summary
TL;DR - GBSK is a large-scale density-clustering algorithm published in IEEE TPAMI 2026 that extracts a sparse “geometric density skeleton” instead of processing every sample directly. It reduces quadratic density-clustering costs to near-linear complexity while retaining competitive accuracy on datasets containing up to 100 million samples.
- GBSK uses granular-ball density statistics as a lightweight proxy for kernel density estimation under balanced granular-ball constraints.
- A divide-and-conquer pipeline identifies mode candidates from random subsamples, aggregates them, and estimates a distribution skeleton for fast clustering.
- Tests across nine datasets showed order-of-magnitude speedups; MNIST8M and the 100-million-sample AGC100M were clustered within minutes on a standard workstation.
- The paper, code, supplementary materials, data, and experiment records are publicly available in the project’s GitHub repository.
Sources (1)
顶刊TPAMI 2026！GBSK：面向复杂分布的大规模高效聚类
TL;DR - GBSK is a large-scale density-clustering algorithm published in IEEE TPAMI 2026 that extracts a sparse “geometric density skeleton” instead of processing every sample directly. It reduces quadratic density-clustering costs to near-linear complexity while retaining competitive accuracy on datasets containing up to 100 million samples.
- GBSK uses granular-ball density statistics as a lightweight proxy for kernel density estimation under balanced granular-ball constraints.
- A divide-and-conquer pipeline identifies mode candidates from random subsamples, aggregates them, and estimates a distribution skeleton for fast clustering.
- Tests across nine datasets showed order-of-magnitude speedups; MNIST8M and the 100-million-sample AGC100M were clustered within minutes on a standard workstation.
- The paper, code, supplementary materials, data, and experiment records are publicly available in the project’s GitHub repository.