Files in this item



application/pdfYIN-THESIS-2020.pdf (559kB)Restricted Access
(no description provided)PDF


Title:Belief propagation on factor graph neural networks
Author(s):Yin, Jialong
Advisor(s):Koyejo, Sanmi
Department / Program:Mechanical Sci & Engineering
Discipline:Mechanical Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Belief Propagation
Graph Neural Networks
Abstract:Probabilistic graphical models are a statistical framework of conditional dependent random variables in the form of graph. A traditional method to perform inference over these random variables is Belief Propagation. It can compute an exact solution for non-loopy factor graphs. However, when provided with loopy factor graphs, it only estimates querying probabilities approximately. In this paper, we propose a Graph Neural Networks (GNN) based on the message passing mechanism of Belief Propagation. Representations and functions in Belief Propagation are fully learned in GNN. We apply our idea to three experiments: inference on loopy factor graphs, error correction decoding task and point cloud segmentation. All the results show our proposal has promising performance compared to the state of arts in each task. Learned representations and functions can also generalize to the factor graphs with different size and structure.
Issue Date:2020-07-24
Rights Information:Copyright 2020 Jialong Yin
Date Available in IDEALS:2020-10-07
Date Deposited:2020-08

This item appears in the following Collection(s)

Item Statistics