ARO Statistical Foundations for Analyzing Large Collections of Network-Data Objects
Abstract
In this work we investigated new ways to use concepts from geometry, probability and statistics on manifolds in order to analyze and categorize classes of data objects. This project progressed along the following three thrusts: 1. It associated classes of data objects with submanifolds and singular quotient spaces of Euclidean spaces through their adjacency matrices and (combinatorial) Laplacians. Techniques of differential geometry enabled mathematical characterization of these novel spaces and of appropriate notions of averaging of objects in these spaces. 2. Concepts from probability and statistics on manifolds were extended to adapt to the high-dimensional and geometrically complex nature of these data spaces. An appropriate probabilistic framework was developed for describing the statistical behavior of such averages. 3. From this probabilistic foundation, a variety of statistical methodologies were constructed and tested for analyzing large collections of data objects.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 31, 2018
- Accession Number
- AD1068368
Entities
People
- Eric D. Kolaczyk
Organizations
- Boston University