Fusion of Graph Neural Networks via Optimal Transport
Abstract
In this paper, we explore the idea of combining GCNs into one model. To that end, we align the weights of different models layer-wise using optimal transport (OT). We present and evaluate three types of transportation costs and show that the studied fusion method consistently outperforms the performance of vanilla averaging. Finally, we present results suggesting that model fusion using OT is harder in the case of GCNs than MLPs and that incorporating the graph structure into the process does not improve the performance of the method.
Cite
Text
Ormaniec et al. "Fusion of Graph Neural Networks via Optimal Transport." ICLR 2025 Workshops: WSL, 2025. doi:10.48550/arxiv.2503.21579Markdown
[Ormaniec et al. "Fusion of Graph Neural Networks via Optimal Transport." ICLR 2025 Workshops: WSL, 2025.](https://mlanthology.org/iclrw/2025/ormaniec2025iclrw-fusion/) doi:10.48550/arxiv.2503.21579BibTeX
@inproceedings{ormaniec2025iclrw-fusion,
title = {{Fusion of Graph Neural Networks via Optimal Transport}},
author = {Ormaniec, Weronika and Vollenweider, Michael and Hoskovec, Elisa},
booktitle = {ICLR 2025 Workshops: WSL},
year = {2025},
doi = {10.48550/arxiv.2503.21579},
url = {https://mlanthology.org/iclrw/2025/ormaniec2025iclrw-fusion/}
}