English

Partitioned K-nearest neighbor local depth for scalable comparison-based learning

Data Structures and Algorithms 2021-12-06 v3 Combinatorics Probability

Abstract

A triplet comparison oracle on a set SS takes an object xSx \in S and for any pair {y,z}S{x}\{y, z\} \subset S \setminus \{x\} declares which of yy and zz is more similar to xx. Partitioned Local Depth (PaLD) supplies a principled non-parametric partitioning of SS under such triplet comparisons but needs O(n2logn)O(n^2 \log{n}) oracle calls and O(n3)O(n^3) post-processing steps. We introduce Partitioned Nearest Neighbors Local Depth (PaNNLD), a computationally tractable variant of PaLD leveraging the KK-nearest neighbors digraph on SS. PaNNLD needs only O(nKlogn)O(n K \log{n}) oracle calls, by replacing an oracle call by a coin flip when neither yy nor zz is adjacent to xx in the undirected version of the KK-nearest neighbors digraph. By averaging over randomizations, PaNNLD subsequently requires (at best) only O(nK2)O(n K^2) post-processing steps. Concentration of measure shows that the probability of randomization-induced error δ\delta in PaNNLD is no more than 2eδ2K22 e^{-\delta^2 K^2}.

Keywords

Cite

@article{arxiv.2108.08864,
  title  = {Partitioned K-nearest neighbor local depth for scalable comparison-based learning},
  author = {Jacob D. Baron and R. W. R. Darling and J. Laylon Davis and R. Pettit},
  journal= {arXiv preprint arXiv:2108.08864},
  year   = {2021}
}

Comments

27 pages, 2 figures

R2 v1 2026-06-24T05:15:54.630Z