Home
Resource
Neural logical reasoning models for graphs and vision
PDF
PDF format is widely accepted and good for printing.
Plug-in required
PDF-1(585.20 kb)
Citation & Export
View Usage Statistics
Staff View

Citation & Export
Hide

Simple citation

Velaga, Pavan Kumar. Neural logical reasoning models for graphs and vision. Retrieved from https://doi.org/doi:10.7282/t3-4f17-5j81

Export

Click here for information about Citation Management Tools at Rutgers.

Statistics
Hide

Description

TitleNeural logical reasoning models for graphs and vision
NameVelaga, Pavan Kumar (author); Pavlovic, Vladimir (chair); Wang, Hao (member); Rutgers University; School of Graduate Studies
Date Created2022
Other Date2022-10 (degree)
SubjectComputer science, Graph reasoning, Neural logical models, Visual analogy reasoning
Extent1 online resource (43 pages) : illustrations
DescriptionOver the years, we have seen the development and success of modern deep learningmodels, which learn the patterns in data to solve a variety of challenging problems. However, these models ignore the reasoning aspect while solving these tasks. Since reasoning is one of the important cognitive abilities that humans use to solve problems, it is intuitive to embed reasoning into deep learning models. After the introduction of differentiable logic reasoning modules in Neural Logic Reasoning, we believe incorporating reasoning modules into existing deep learning architectures would improve their performance in solving tasks where cognition is important. In this thesis, we intend to demonstrate the effectiveness of using a reasoning module in solving graph classification and visual analogy tasks. To solve the graph classification task, we introduce Neural Logic Graph Reasoning (NLGR), where we learn nodes as logical variables and graph as a logical expression. We were successful in improving the performance of some traditional graph neural network models by incorporating logical reasoning modules into them. We also introduced the Neural Logic Visual Reasoning (NLVR) model to solve the visual analogy problems, where we were able to improve baseline models by adding logical reasoning modules. This work reaffirms the utility of reasoning modules in modern deep learning models.
NoteM.S.
NoteIncludes bibliographical references
Genretheses
Persistent URLhttps://doi.org/doi:10.7282/t3-4f17-5j81
LanguageEnglish
CollectionSchool of Graduate Studies Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.
Version 8.5.5
Rutgers University Libraries - Copyright ©2024