English

Uniform spaces and the Newtonian structure of (big)data affinity kernels

General Topology 2017-01-16 v1 Analysis of PDEs

Abstract

Let XX be a (data) set. Let K(x,y)>0K(x,y)>0 be a measure of the affinity between the data points xx and yy. We prove that KK has the structure of a Newtonian potential K(x,y)=φ(d(x,y))K(x,y)=\varphi(d(x,y)) with φ\varphi decreasing and dd a quasi-metric on XX under two mild conditions on KK. The first is that the affinity of each xx to itself is infinite and that for xyx\neq y the affinity is positive and finite. The second is a quantitative transitivity; if the affinity between xx and yy is larger than λ>0\lambda>0 and the affinity of yy and zz is also larger than λ\lambda, then the affinity between xx and zz is larger than ν(λ)\nu(\lambda). The function ν\nu is concave, increasing, continuous from R+\mathbb{R}^+ onto R+\mathbb{R}^+ with ν(λ)<λ\nu(\lambda)<\lambda for every λ>0\lambda>0.

Keywords

Cite

@article{arxiv.1701.03746,
  title  = {Uniform spaces and the Newtonian structure of (big)data affinity kernels},
  author = {Hugo Aimar and Ivana Gómez},
  journal= {arXiv preprint arXiv:1701.03746},
  year   = {2017}
}

Comments

7 pages