Back to search

Clemson University

Efficient Analytics for Large Implicit Correlation Networks

Abstract

dc:description.abstract

Large “correlation”-type networks have applications in machine learning, data mining and analysis of large biological networks. These networks contain short length-k (on the order of thousands) feature vectors for every node in the network that describe the characteristic or behavior of that node. Edges between nodes are expressed as correlations, partial correlations or coherence between the corresponding feature vectors. If the number of nodes n is too large (on the order of millions), then the dense correlation network with size n×n is computationally infeasible even to explicitly store in memory, let alone to use for computation of standard network analytics like clustering or centrality. In this dissertation, we provide an algorithmic framework to analyze large correlation networks efficiently in an implicit fashion, without writing down every entry of the network. Our framework allows us to compute various network analytics like clustering and centrality using spectral techniques (based on linear algebra). It can accommodate several more sophisticated edge weight measurements beyond just correlation, including partial correlation and coherence. Since correlation networks can contain negative edge weights, one of the main innovations of our work is a collection of transformations that re-map node feature vectors so as to produce a non-negative network in which edge weights are approximately the squares of the original weights. We also develop a number of other useful algorithmic tools (e.g., implicit calculation of shrinkage estimators) that play an important role along our implicit computational pipeline.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy (PhD)
Level thesis:degree_level
Dissertation
Discipline thesis:degree_discipline
Computer Science

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Mohan, Raghuveer
Contributors dc:contributor
  • Dr. Brian Dean, Committee Chair
  • Dr. Wayne Goddard
  • Dr. Ilya Safro
  • Dr. Feng Luo

Identifiers

dc:identifier.*
Repository record dc:identifier
https://open.clemson.edu/all_dissertations/1826

Chain of custody

source
Harvested from
Clemson University
Base URL
open.clemson.edu/do/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Mohan, Raghuveer. Efficient Analytics for Large Implicit Correlation Networks. Dissertation thesis, https://open.clemson.edu/all_dissertations/1826