Application of Nonparametric regression to Big data in Astroinformatics

O. Kounchev,
Institute of Mathematics and Informatics, Bulgarian Academy of Sciences

Abstract.

  In this talk we present new models for univariate and multivariate
nonparametric regressions based on smoothing L-splines and smoothing polysplines,
respectively.
  These multivariate models enjoy very fast algorithms for computations unlike the
competitive approaches as Kriging, Thin plate splines, Radial Basis functions, etc.
Another essential feature of the new nonparametric regressions is the existence of fast
computation algorithms of the smoothing parameter by means of Cross-validation (or GCV), and of
"confidence intervals" for the models.
  We present some possible applications to data arising in Astronomy and other
areas, which have been traditionally treated by other approaches.

The present research is joint with Ts. Tsachev, H. Render, and supported by
DH 02-13 and I 02-19 of Bulgarian NSF.

References
1. Eric D. Feigelson, G. Jogesh Babu, ​Modern Statistical Methods for Astronomy: With
R Applications. Cambridge UP.
2. Yu Yu, Jun Zhang, Yipeng Jing, and Pengjie Zhang, Kriging, interpolating cosmic
velocity field, Phys. Rev. D 92, 083527 – Published 27 October 2015