FOSC-X finds top-M optimal clusterings from hierarchies in linear time
Alternative cluster structures usually missed by single-solution methods are now discoverable at scale.
Researchers Connor Simpson and Ricardo Campello have released FOSC-X, a framework that extracts the top-M globally optimal flat clusterings from hierarchical cluster trees. Unlike conventional methods that return a single optimal cut, FOSC-X uses dynamic programming to find multiple high-quality solutions, each capturing different aspects of the data's hierarchical structure. Without cluster-count constraints, the algorithm runs in polynomial time by combining locally optimal partial candidates within subtrees. When users need to limit the number of clusters (e.g., to avoid meaningless granularity), FOSC-X maintains compact sets of feasible candidates using lower and upper bounds, pruning infeasible or dominated combinations. This guarantees an optimal ranking of the top-M solutions with linear-time complexity in both the number of cluster nodes and dataset size.
FOSC-X addresses a critical gap in cluster analysis: the inability to explore alternative groupings from a single hierarchy. Experiments demonstrate that the framework efficiently uncovers structures that single-solution extraction methods miss, all while maintaining computational efficiency. For practitioners, this means no more manually fiddling with cut thresholds or rerunning different clustering algorithms to find diverse, meaningful partitions. The framework is particularly useful for domains where interpretability and pluralistic views of data are essential, such as bioinformatics, customer segmentation, or anomaly detection. The paper is available on arXiv (2606.18972) and includes implementations that can be integrated into existing clustering pipelines.
- FOSC-X extracts top-M optimal flat clusterings from hierarchical trees using dynamic programming in linear time.
- Users can enforce constraints on cluster count, and the algorithm still guarantees optimal rankings with linear complexity.
- Experiments show it reveals alternative clustering structures overlooked by traditional single-solution extraction methods.
Why It Matters
FOSC-X gives data scientists multiple high-quality clusterings from one hierarchy, improving exploration and interpretability without extra computational cost.