Toward Compositional Generalization with Neuro-Symbolic AI: A Comprehensive Overview - Archive ouverte HAL
Communication Dans Un Congrès Année : 2024

Toward Compositional Generalization with Neuro-Symbolic AI: A Comprehensive Overview

Résumé

This review emphasizes the fascinating convergence of Neuro-Symbolic AI (NeSy) and Compositional Generalization (CoGe), examining how these models might potentially transform AI by enabling real human-like intelligence. This research contends that NeSy's capacity to combine the advantages of neural and symbolic techniques has enormous potential for addressing the CoGe dilemma. CoGe necessitates the ability to learn and use information in unexpected settings through the flexible assembly of existing building components. NeSy architectures, with their distinct combination of symbolic reasoning and flexible learning, provide a viable option for overcoming this critical hurdle. In this paper, we highlighted some of the most important concepts of both NeSy and CoGe, showcasing the cutting-edge research trends shaping these fields, we delve into their diverse techniques and methods. Drawing upon cognitive science studies and concrete AI-based works, we illustrate the multitude of possibilities for implementing CoGe within NeSy. Finally, we discuss the results performed by these studies, their commonalities, we then present our proposition and address the open challenges that lie ahead on the path towards true CoGe with NeSy.
Fichier principal
Vignette du fichier
EDIS 2024_IEEExplorer_version.pdf (565.77 Ko) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04705213 , version 1 (22-09-2024)

Identifiants

  • HAL Id : hal-04705213 , version 1

Citer

Nassim Lateb, Farida Bouarab-Dahmani, Florence Sèdes. Toward Compositional Generalization with Neuro-Symbolic AI: A Comprehensive Overview. 4th IEEE international Conference on Embedded & Distributed Systems (EDIS4 2024), University of Bechar-Tahri Mohamed, Bechar, Algeria, Nov 2024, Bechar, Algeria. à paraître. ⟨hal-04705213⟩
0 Consultations
0 Téléchargements

Partager

More