From 54355bfc3c6c5b4997a4885c2843365be833abea Mon Sep 17 00:00:00 2001 From: ValentinaSanchezMelchor Date: Mon, 18 May 2026 22:12:09 +0200 Subject: [PATCH 1/2] Add DHGNN challenge submission draft --- 2026_tdl_challenge/submissions/dhgnn/README.md | 12 ++++++++++++ 1 file changed, 12 insertions(+) create mode 100644 2026_tdl_challenge/submissions/dhgnn/README.md diff --git a/2026_tdl_challenge/submissions/dhgnn/README.md b/2026_tdl_challenge/submissions/dhgnn/README.md new file mode 100644 index 000000000..c3c337238 --- /dev/null +++ b/2026_tdl_challenge/submissions/dhgnn/README.md @@ -0,0 +1,12 @@ +# DHGNN Challenge Submission Draft + + This folder tracks the draft Track 2 submission for DHGNN, based on: + + Ma et al., "Directed Hypergraph Representation Learning for Link Prediction," AISTATS 2024. + + Paper: https://proceedings.mlr.press/v238/ma24b/ma24b.pdf + + Official implementation: https://github.com/mazitong/DHGNN + + This is an exploratory draft. The first implementation step is to verify how the original link-prediction model maps to + the TopoBench/GraphUniverse challenge pipeline. \ No newline at end of file From dc507956c8409116398e78b9900b28c4e9d34705 Mon Sep 17 00:00:00 2001 From: ValentinaSanchezMelchor Date: Tue, 19 May 2026 20:46:07 +0200 Subject: [PATCH 2/2] Switch draft submission to SheafHyperGNN --- .../submissions/dhgnn/README.md | 12 ----- .../submissions/sheaf_hypergnn/README.md | 49 +++++++++++++++++++ 2 files changed, 49 insertions(+), 12 deletions(-) delete mode 100644 2026_tdl_challenge/submissions/dhgnn/README.md create mode 100644 2026_tdl_challenge/submissions/sheaf_hypergnn/README.md diff --git a/2026_tdl_challenge/submissions/dhgnn/README.md b/2026_tdl_challenge/submissions/dhgnn/README.md deleted file mode 100644 index c3c337238..000000000 --- a/2026_tdl_challenge/submissions/dhgnn/README.md +++ /dev/null @@ -1,12 +0,0 @@ -# DHGNN Challenge Submission Draft - - This folder tracks the draft Track 2 submission for DHGNN, based on: - - Ma et al., "Directed Hypergraph Representation Learning for Link Prediction," AISTATS 2024. - - Paper: https://proceedings.mlr.press/v238/ma24b/ma24b.pdf - - Official implementation: https://github.com/mazitong/DHGNN - - This is an exploratory draft. The first implementation step is to verify how the original link-prediction model maps to - the TopoBench/GraphUniverse challenge pipeline. \ No newline at end of file diff --git a/2026_tdl_challenge/submissions/sheaf_hypergnn/README.md b/2026_tdl_challenge/submissions/sheaf_hypergnn/README.md new file mode 100644 index 000000000..37b5b0d73 --- /dev/null +++ b/2026_tdl_challenge/submissions/sheaf_hypergnn/README.md @@ -0,0 +1,49 @@ +# Track 2 — SheafHyperGNN Submission + +## Track + +Track 2 — Topological Neural Networks (TNNs) + +## Team Name + +s/pairwise/ho + +## Model + +Sheaf Hypergraph Networks (SheafHyperGNN) + +## Status + +Draft / work in progress + +## Summary + +This draft PR develops a TopoBench-native implementation of SheafHyperGNN from +"Sheaf Hypergraph Networks" (Duta et al., NeurIPS 2023) for the 2026 TDL +Challenge. + +The model generalises standard hypergraph convolutions by equipping every node +and hyperedge with a *d*-dimensional stalk and learning per-pair restriction +maps that define a cellular sheaf over the hypergraph. The resulting sheaf +Laplacian replaces the standard hypergraph Laplacian in the diffusion operator, +enabling richer structural representations without additional data requirements +beyond the incidence matrix. + +## Planned Implementation + +- [x] Inspect official implementation and paper equations; confirm feasibility. +- [ ] Add SheafHyperGNN backbone under `topobench/nn/backbones/hypergraph/sheaf_hypergnn.py`. +- [ ] Add Hydra config under `configs/model/hypergraph/sheaf_hypergnn.yaml`. +- [ ] Add unit tests. +- [ ] Update `test/pipeline/test_pipeline.py`. +- [ ] Run the official GraphUniverse evaluation notebook. +- [ ] Add generated `results.json`. + +## Reference + +Duta, I., Cassarà, G., Silvestri, F., & Liò, P. +"Sheaf Hypergraph Networks." *NeurIPS 2023.* + +Paper: https://arxiv.org/abs/2309.17116 + +Official implementation: https://github.com/IuliaDuta/sheaf_HNN