Statistical models of graph and relational data from probabilistic symmetries
Abstract
In this proposal, the two PIs propose to investigate statistical and Turing computational aspects of variational inference in exchangeable models for arrays of statistical data that can be represented by graphs, matrices or arrays and will develop new methods that are beyond the exchangeable case. The research will focus on (1) models based on exchangeable distributions; and (2) models for sparse graphs and matrices. This work will extend the current notion of exchangeable processes that have successfully led to Hierarchical Dirichlet Allocation and Hierarchical Bayesian models for modeling and making inference on various types of datasets.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 22, 2016
- Source ID
- FA95501510074
Entities
People
- Peter Orbanz
Organizations
- Air Force Office of Scientific Research
- Trustees of Columbia University in the City of New York
- United States Air Force