English

A Generalization of Self-Improving Algorithms

Computational Geometry 2020-08-24 v2 Computational Complexity

Abstract

Ailon et al. [SICOMP'11] proposed self-improving algorithms for sorting and Delaunay triangulation (DT) when the input instances x1,,xnx_1,\cdots,x_n follow some unknown \emph{product distribution}. That is, xix_i comes from a fixed unknown distribution Di\mathsf{D}_i, and the xix_i's are drawn independently. After spending O(n1+ε)O(n^{1+\varepsilon}) time in a learning phase, the subsequent expected running time is O((n+H)/ε)O((n+ H)/\varepsilon), where H{HS,HDT}H \in \{H_\mathrm{S},H_\mathrm{DT}\}, and HSH_\mathrm{S} and HDTH_\mathrm{DT} are the entropies of the distributions of the sorting and DT output, respectively. In this paper, we allow dependence among the xix_i's under the \emph{group product distribution}. There is a hidden partition of [1,n][1,n] into groups; the xix_i's in the kk-th group are fixed unknown functions of the same hidden variable uku_k; and the uku_k's are drawn from an unknown product distribution. We describe self-improving algorithms for sorting and DT under this model when the functions that map uku_k to xix_i's are well-behaved. After an O(poly(n))O(\mathrm{poly}(n))-time training phase, we achieve O(n+HS)O(n + H_\mathrm{S}) and O(nα(n)+HDT)O(n\alpha(n) + H_\mathrm{DT}) expected running times for sorting and DT, respectively, where α()\alpha(\cdot) is the inverse Ackermann function.

Keywords

Cite

@article{arxiv.2003.08329,
  title  = {A Generalization of Self-Improving Algorithms},
  author = {Siu-Wing Cheng and Man-Kwun Chiu and Kai Jin and Man Ting Wong},
  journal= {arXiv preprint arXiv:2003.08329},
  year   = {2020}
}
R2 v1 2026-06-23T14:18:56.842Z