Szczegóły publikacji
Opis bibliograficzny
Evolutionary design of Graph Neural Network architectures for graph classification / Maciej KRZYWDA, Szymon Łukasik, Amir H. Gandomi // W: Progress in Polish artificial intelligence research 6 [Dokument elektroniczny] : 6th Polish Conference on Artifical Intelligence (PP-RAI'2025) : 07–09.04.2025, Katowice, Poland / ed. by Rafał Doroz, Beata Zielosko. — Wersja do Windows. — Dane tekstowe. — Katowice : Wydawnictwo Uniwersytetu Śląskiego, 2025. — e-ISBN: 978-83-226-4405-8. — S. 247–253. — Wymagania systemowe: Adobe Reader. — Bibliogr. s. 252–253, Abstr. — Sz. Łukasik - afiliacja: Systems Research Institute, Polish Academy of Sciencesul, Warsaw, Poland
Autorzy (3)
- AGHKrzywda Maciej
- Łukasik Szymon
- Gandomi Amir H.
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 165797 |
|---|---|
| Data dodania do BaDAP | 2026-02-03 |
| Tekst źródłowy | URL |
| Rok publikacji | 2025 |
| Typ publikacji | fragment monografii pokonferencyjnej |
| Otwarty dostęp |  |
| Creative Commons | |
| Wydawca | Uniwersytet Śląski w Katowicach |
Abstract
Graph Neural Networks (GNNs) are a family of neural archi-tectures operating on graphs, inspired by interactions between nodes ona graph. This study develops the first stage of a method that uses evolutionary algorithms to design and optimize graph convolutional neural networksfor graph classification. Our method allows us to set critical parameters,such as batch size, number of hidden channels, and the choice and parameters of optimizer and loss function. The approach uses multi-criteria designand can be applied to the architectural-level hyperparameter tuning of graphneural networks, such as Graph Attention Networks (GAT), Graph Isomor-phism Networks (GIN), and Graph Convolutional Networks (GCN). We evaluate the proposed method on benchmark datasets and observe encouragingperformance from the synthesized networks.
