Answers are grounded in this paper and cite their sources.
Theoretical analysis reveals hyperoctahedral symmetry preservation in grid-based density clustering, highlighting opportunities for canonical indexing and dataset deduplication.
Grid-based density clustering methods—FlowGrid, FLOPTICS, and grid-accelerated DBSCAN and OPTICS—partition a d-dimensional feature space into an md array of bins and group the non-empty bins by density reachability. They leave a symmetry unexploited: relabelling and reflecting the feature axes permutes the bins while preserving density. We formalise this as an action of the hyperoctahedral group Bd = C2 \( \) Sd of order 2dd! on bin-occupancy functions. We prove that the action preserves the Chebyshev bin-adjacency underlying grid reachability (Theorem 1); that grid density clustering is Bd-equivariant (Theorem 2); that configurations up to symmetry are counted by a Burnside average over Bd (Theorem 3), with an explicit parity-split closed form for the octahedral case at every resolution, which we have not found recorded elsewhere (Theorem 4 and Corollary 2); and that a canonical-form algorithm computes a unique representative per orbit in O (2dd! md) time (Theorem 5). We are explicit about scope: a generic dataset has a trivial stabiliser, so there is no per-run gain; the benefit is canonical indexing, deduplication of symmetry-closed libraries, and caching when the same measurement recurs under a different axis convention. The group specialises to the D4 and Oh symmetries of the DR Code. Symmetry enters twice: as the group acting on the grid, and as the line between symmetric instances and the generic asymmetric ones. All theorems and enumeration formulae are confirmed computationally.
No takes yet. Share an insight, caveat, or question.
Sriphum et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: