GCN โ Cora citation network
A 2-layer Graph Convolutional Network (Kipf & Welling 2017) trained on the Cora citation graph for 7-class node classification. Released as xaitalk's cross-framework XAI demo on graph architectures.
Files
| File | Format | Size |
|---|---|---|
gcn_cora.pt |
PyTorch state_dict | ~110 KB |
Architecture
| Property | Value |
|---|---|
| Layers | 2 (graph convolution + dropout + graph convolution) |
| Hidden dim | 16 |
| Input features | 1433 (word-vector features) |
| Output | 7-class logits |
| Parameters | ~23 K |
Standard GCN (Kipf 2017). Test accuracy: 81.7% on the canonical Cora split (140 train / 500 val / 1000 test nodes).
Cross-framework verification
These weights are validated by xaitalk's gnn benchmark on Cora
(19 methods):
| Methods | Passing at r โฅ 0.95 | Min(min_r) | Verified |
|---|---|---|---|
| 19 | 19/19 | 0.9756 | 2026-05-09 |
The min_r=0.9756 sits on the stochastic vargrad method
(variance-of-gradients) at default n_samples=50. All deterministic
methods (full gradient family, full LRP family, DeepLIFT, GradCAM) at
r โฅ 0.999 cross-framework.
Usage
from xaitalk.hub import ensure_model
import torch
ckpt_path = ensure_model('gnn/gcn-cora')
# Architecture class lives in xaitalk
from xaitalk.models import GCN
model = GCN(in_features=1433, hidden=16, num_classes=7)
model.load_state_dict(torch.load(ckpt_path, weights_only=True))
model.eval()
# Run XAI on a Cora subgraph
import xaitalk
expl = xaitalk.explain(model, x, method='lrp_epsilon', target_class=2)
Training data
Cora citation network โ 2708 papers, 7 classes (machine-learning subfields), 5429 citation edges, word-vector features per paper.
License
Apache 2.0.
Citation
@inproceedings{kipf2017gcn,
author = {Kipf, Thomas N. and Welling, Max},
title = {Semi-Supervised Classification with Graph Convolutional Networks},
booktitle = {International Conference on Learning Representations (ICLR)},
year = {2017}
}
@software{paul2026xaitalk,
author = {Paul, Alexander},
title = {xaitalk: Cross-Framework Explainable AI Library},
year = {2026},
url = {https://xaitalk.com}
}
Links
- xaitalk website: https://xaitalk.com
- Framework GitHub: https://github.com/alexanderfpaul/xaitalk-framework
- GNN comparison script:
examples/comparison/run_gnn_3framework_comparison.py