Universally Useful Primitives for Aligning Networks Across Time and Space
Abstract
Their team has, in the course of working on the DARPA Modeling Adversarial Activity (MAA) program, further developed a fast, flexible suite of graph matching tools designed to robustly align large networks in the presence of noise, paying special heed to developing methods for multiplex matching; developed the theory and methodology behind Graph Matching Matched Filters which provide a principled, scalable method for discovering noisy subgraphs in a larger background graph; provided an open source R code-base, denoted iGraphMatch, for implementing our graph matching and graph matching matched filters methods and their competitors at scale; further developed the theory of vertex nomination, developing the analogues of the classical statistical concepts of consistency and Bayes optimality in the context of vertex nomination; developed a novel concept of adversarial contamination and data-adaptive regularization inthe context of vertex nomination; developed a suite of flexible vertex nomination algorithms designed to be implemented on large, noisy networks; produced illustrative simulations and data analyses on MAA provided data and on externally provided real data sources.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 2020
- Accession Number
- AD1101398
Entities
People
- Carey E. Priebe
- Daniel L. Sussman
- Vincent Lyzinski
- Youngser Park
Organizations
- University of Massachusetts Amherst