Graph Neural Networks Meet Neural-Symbolic Computing: A Survey and\n Perspective

2020113 citationsPreprintgreen Open Access

Authors

Luís C. Lamb · Universidade Federal do Rio Grande do Sul

Artur d’Avila Garcez · City, University of London

Marco Gori · University of Siena

Marcelo Prates · Universidade Federal do Rio Grande do Sul

Pedro H. C. Avelar · University of Siena

Moshe Y. Vardi · Rice University

Abstract

Neural-symbolic computing has now become the subject of interest of both\nacademic and industry research laboratories. Graph Neural Networks (GNN) have\nbeen widely used in relational and symbolic domains, with widespread\napplication of GNNs in combinatorial optimization, constraint satisfaction,\nrelational reasoning and other scientific domains. The need for improved\nexplainability, interpretability and trust of AI systems in general demands\nprincipled methodologies, as suggested by neural-symbolic computing. In this\npaper, we review the state-of-the-art on the use of GNNs as a model of\nneural-symbolic computing. This includes the application of GNNs in several\ndomains as well as its relationship to current developments in neural-symbolic\ncomputing.\n

Topics & Keywords

Advanced Graph Neural Networks Neural Networks and Applications Bayesian Modeling and Causal Inference

UN Sustainable Development Goals

Industry, innovation and infrastructure

Publication Details

Published in: arXiv (Cornell University)

DOI: 10.48550/arxiv.2003.00330