Parallelizing Node-Level Explainability in Graph Neural Networks Journal: Neural Networks
Oscar Llorente, Jaime Boal, Eugenio F. Sánchez-Úbeda, Antonio Diaz-Cano, Miguel Familiar · 2026
Read Parallelizing Node-Level Explainability in Graph Neural Networks Journal: Neural Networks by Oscar Llorente, Jaime Boal, Eugenio F. Sánchez-Úbeda, Antonio Diaz-Cano, Miguel Familiar (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-04807.txt.
Description
Explaining why a graph neural network labels a node can take longer than the prediction itself on large graphs. Oscar Llorente and 4 co-authors tackle that bottleneck in a 2026 paper. Graph neural networks use connections between items to classify nodes, find links, and label whole graphs. Node-level explanations, though, grow extremely time-consuming as graphs expand, and batching often weakens explanation quality.
Their approach partitions the graph, splitting it into disjoint subgraphs that can be processed in parallel. Each part's explanations are computed separately, then combined. Faster explanations make these models easier to audit in settings like fraud or biology networks, where users need reasons, not just labels, at realistic scale.
Catalog record
| CREATOR | Oscar Llorente, Jaime Boal, Eugenio F. Sánchez-Úbeda, Antonio Diaz-Cano, Miguel Familiar |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 68 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-04807