Publicação
FuzzyKG
| Resumo: | This paper introduces FuzzyKG, a neuro-symbolic framework for extracting interpretable symbolic knowledge from fuzzy neural networks. FuzzyKG transforms fuzzy rules learned by Gaussian membership functions and logical neurons into human-readable fuzzy axioms and organizes them into a structured Knowledge Graph (KG), explicitly representing relationships between features, linguistic terms, rule antecedents, and consequents. The framework preserves the numerical semantics of the underlying fuzzy model—such as feature relevance, rule strength, and class polarity—while exposing them in a symbolic, graph-based representation. Experimental results on benchmark datasets show that FuzzyKG maintains competitive predictive performance while producing coherent axioms and meaningful graph structures, enabling transparent analysis and supporting future neuro-symbolic extensions. |
|---|---|
| Autores principais: | de Campos Souza, Paulo Vitor |
| Outros Autores: | Rosenfeld, Liah |
| Assunto: | Neuro-Symbolic Fuzzy Neural-Networks Explainability Knowledge-Graphs Interpretability Control and Optimization Logic Modelling and Simulation |
| Ano: | 2026 |
| País: | Portugal |
| Tipo de documento: | documento de conferência |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | This paper introduces FuzzyKG, a neuro-symbolic framework for extracting interpretable symbolic knowledge from fuzzy neural networks. FuzzyKG transforms fuzzy rules learned by Gaussian membership functions and logical neurons into human-readable fuzzy axioms and organizes them into a structured Knowledge Graph (KG), explicitly representing relationships between features, linguistic terms, rule antecedents, and consequents. The framework preserves the numerical semantics of the underlying fuzzy model—such as feature relevance, rule strength, and class polarity—while exposing them in a symbolic, graph-based representation. Experimental results on benchmark datasets show that FuzzyKG maintains competitive predictive performance while producing coherent axioms and meaningful graph structures, enabling transparent analysis and supporting future neuro-symbolic extensions. |
|---|
Atividades financiadas
Não foram encontrados projetos financiados.