PIRSA:26060015

Video URL

https://pirsa.org/26060015

Harnessing information from higher order statistics in cosmology - k-nearest neighbor (kNN) distributions

Banerjee, A. (2026). Harnessing information from higher order statistics in cosmology - k-nearest neighbor (kNN) distributions. Perimeter Institute for Theoretical Physics. https://pirsa.org/26060015

Banerjee, Arka. Harnessing information from higher order statistics in cosmology - k-nearest neighbor (kNN) distributions. Perimeter Institute for Theoretical Physics, Jun. 02, 2026, https://pirsa.org/26060015 

          @misc{ scivideos_PIRSA:26060015,
            doi = {10.48660/26060015},
            url = {https://pirsa.org/26060015},
            author = {Banerjee, Arka},
            keywords = {Cosmology},
            language = {en},
            title = {Harnessing information from higher order statistics in cosmology - k-nearest neighbor (kNN) distributions},
            publisher = {Perimeter Institute for Theoretical Physics},
            year = {2026},
            month = {jun},
            note = {PIRSA:26060015 see, \url{https://scivideos.org/pirsa/26060015}}
          }

Arka Banerjee Indian Institute of Science Education and Research Pune

Talk numberPIRSA:26060015
DOI10.48660/26060015
Source RepositoryPIRSA
Collection
Talk Type Scientific Series
Subject

Abstract

Current and upcoming cosmological surveys will map the observable Universe with increasingly greater precision. To extract the maximum information about cosmology from these surveys, especially from smaller scales, it is imperative to move beyond traditional 2-point analyses. In this talk, I will introduce a new set of summary statistics, the k-Nearest Neighbor (kNN) distributions, which are sensitive to moments of all N-point functions in the data, while computationally scaling like 2-point measurements. I will discuss how these summary statistics can measure auto and cross-correlations in both discrete and continuous datasets, as well as their connections to other higher-order statistics proposed in the literature. I will outline various science cases where these statistics can be applied to either increase the significance of detection or extract tighter constraints on parameters of interest. Finally, I will discuss attempts to model these statistics using ingredients that have already been successfully applied to modeling 2-point functions in both real and redshift space.