Track: Track1; Team name: Abdulrahman; Model: Dir-GNN (Directed Graph Neural Network)

[abdulrahman-1212]

Checklist

• My pull request has a clear and explanatory title.
• My pull request passes the Linting test.
• I added appropriate unit tests and I made sure the code passes all unit tests. (refer to comment below)
• My PR follows PEP8 guidelines. (refer to comment below)
• My code is properly documented, using numpy docs conventions, and I made sure the documentation renders properly.
• I linked to issues and PRs that are relevant to this PR.

Dir-GNN: Edge Directionality Improves Learning on Heterophilic Graphs

Paper

• Title: Edge Directionality Improves Learning on Heterophilic Graphs
• Authors: Emanuele Rossi, Bertrand Charpentier, Francesco Di Giovanni, Fabrizio Frasca, Stephan Günnemann, Michael M. Bronstein
• Venue: Learning on Graphs (LoG) Conference, 2024
• ArXiv: https://arxiv.org/abs/2305.10498
• Official code: https://github.com/emalgorithm/directed-graph-neural-network

Track

track-1-gnn

Summary

Dir-GNN extends any Message Passing Neural Network (MPNN) to directed graphs by performing
separate aggregations of incoming and outgoing edges. On heterophilic graphs, treating
edges as directed increases effective homophily, providing significant gains without
architectural complexity.

Key theoretical result: Dir-GNN matches the expressivity of the Directed Weisfeiler-Lehman
test, strictly exceeding conventional MPNNs.

Components added

• topobench/nn/backbones/graph/dir_gnn.py — backbone implementation
• configs/model/graph/dir_gnn.yaml — Hydra config
• test/nn/backbones/graph/test_dir_gnn.py — unit tests
• 2026_tdl_challenge/results.json — GraphUniverse evaluation results