Szczegóły publikacji
Opis bibliograficzny
Symbolic vs black-box explanations: a model-driven approach using grammatical evolution / Dominik SEPIOŁO, Antoni LIGĘZA // W: FedCSIS [Dokument elektroniczny] : proceedings of the 20th conference on Computer Science and Intelligence Systems : September 14–17, 2025, Kraków, Poland / eds. Marek Bolanowski, [et al.]. — Wersja do Windows. — Dane tekstowe. — Warsaw : Polskie Towarzystwo Informatyczne ; [Piscataway] : IEEE, cop. 2025. — ( Annals of Computer Science and Information Systems ; ISSN 2300-5963 ; vol. 43 ). — Dod. ISBN: 979-8-3315-1531-7. — ISBN: 978-83-973291-7-1; e-ISBN: 978-83-973291-6-4. — S. 381–386. — Wymagania systemowe: Adobe Reader. — Tryb dostępu: https://annals-csis.org/Volume_43/pliks/volume_43.pdf [2025-11-04]. — Bibliogr. s. 386, Abstr.
Autorzy (2)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 163988 |
|---|---|
| Data dodania do BaDAP | 2025-11-06 |
| DOI | 10.15439/2025F5371 |
| Rok publikacji | 2025 |
| Typ publikacji | materiały konferencyjne (aut.) |
| Otwarty dostęp |  |
| Creative Commons | |
| Wydawca | Institute of Electrical and Electronics Engineers (IEEE) |
| Konferencja | Conference on Computer Science and Intelligence Systems 2025 |
| Czasopismo/seria | Annals of Computer Science and Information Systems |
Abstract
Black-box explainability tools like LIME and SHAP are widely used to interpret machine learning models. However, their post-hoc, local nature often results in inconsistent and semantically opaque explanations. This paper presents a model-driven explainability approach using grammatical evolution (GE), enabling the discovery of symbolic, human-readable models. We compare black-box explanations to symbolic GE-generated models on two benchmark tasks: a quadratic equation classification problem and the Iris dataset. GE produces interpretable, consistent, and semantically meaningful expressions consistent with domain knowledge, offering a more trustworthy foundation for explainable AI. The use of Meaningful Intermediate Variables (MIVs) further improves the clarity and expressiveness of the symbolic models.
