Citation
TGraphX is described in the arXiv preprint arXiv:2504.03953 (Sajjadi & Eramian, 2025). Use the BibTeX entry or plain-text format below to cite it in academic work. The canonical citation target is the arXiv record.
| Title | TGraphX: Tensor-Aware Graph Neural Network for Multi-Dimensional Feature Learning |
|---|---|
| Authors | Arash Sajjadi, Mark Eramian |
| Affiliation | University of Saskatchewan |
| Year | 2025 |
| arXiv ID | 2504.03953 |
| arxiv.org/pdf/2504.03953 | |
| HTML | arxiv.org/html/2504.03953v1 |
@misc{sajjadi2025tgraphx,
title = {{TGraphX}: Tensor-Aware Graph Neural Network for
Multi-Dimensional Feature Learning},
author = {Sajjadi, Arash and Eramian, Mark},
year = {2025},
eprint = {2504.03953},
archivePrefix= {arXiv},
primaryClass = {cs.LG},
url = {https://arxiv.org/abs/2504.03953}
}
Sajjadi, A., & Eramian, M. (2025). TGraphX: Tensor-Aware Graph Neural Network for Multi-Dimensional Feature Learning. arXiv:2504.03953. https://arxiv.org/abs/2504.03953
The preprint introduces TGraphX and its core design: a PyTorch-based graph learning framework
where node and edge features can be multi-dimensional tensors (e.g., [C, H, W]
or [C, D, H, W])
rather than flat vectors. The paper motivates the tensor-native design,
describes the tgx.Graph
container and message-passing layers, and situates TGraphX within the broader graph learning ecosystem.
The arXiv record is the authoritative citation source. If you are building on a specific version of the software, also note the version (available on PyPI) and the commit hash if reproducibility is important.