Research map / Graph learning / Graph self-supervised learning
Aitchison Embeddings for Learning Compositional Graph Representations
Nikolaos Nakis, Chrysoula Kosma, Panagiotis Promponas, Michail Chatzianastasis, Giannis Nikolentzos
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Aitchison/ILR geometry instead of Euclidean geometry on the simplex (ablation) | AICoG latent distance model · AstroPh link prediction (D=64) | AUC-ROC: .863 -> .976 | Improved |
| use Aitchison/ILR geometry instead of Euclidean geometry on the simplex (ablation) | AICoG latent distance model · Cora node classification (D=64) | Micro-F1: .442 -> .831 | Improved |
| learn the ILR basis jointly instead of fixed Helmert basis (ablation) | AICoG · link prediction and node classification | AUC-ROC / Micro-F1: Cora D=64 .851 -> .852 | No clear effect |
| subcompositional restriction of trained model without retraining (ablation) | AICoG (HB) · Cora link prediction (D=8) | AUC-ROC: .837 -> .807 | No clear effect |
| subcompositional restriction of trained model without retraining (ablation) | AICoG (HB) · Cora node classification (D=8) | Micro-F1: .769 -> .672 | Hurt |
| AICoG instead of MMSBM for membership recovery | synthetic graphs · Bilinear-continuous | l1 distance: 1.456 -> 0.922 | Improved |
| AICoG instead of MMSBM for interior memberships | Cora · Cora memberships | entropy: 0.191 -> 1.064 | Improved |
| AICoG vs Euclidean Node2vec/Role2vec baselines on node classification | AICoG · Cora node classification (D=8) | Micro-F1: .769 vs ROLE2VEC .783 | No clear effect |
| use AICoG instead of SLIM-RAA | AICoG · DBLP link prediction (D=64) | AUC-ROC: .962 vs .962 | No clear effect |
| AICoG vs SLIM-RAA at low dimension | AICoG · Cora node classification (D=8) | Micro-F1: .628 -> .769 | Improved |
Closest papers by meaning
- On the Equivalence between Positional Node Embeddings and Structural Graph Representations (ICLR 2020 · similarity 0.88)
- Capacity and Bias of Learned Geometric Embeddings for Directed Graphs (NeurIPS 2021 · similarity 0.88)
- Weighted Embeddings for Low-Dimensional Graph Representation (AAAI 2025 · similarity 0.87)
- Learning Representations for Hierarchies with Minimal Support (NeurIPS 2024 · similarity 0.87)
- Equivariance Everywhere All At Once: A Recipe for Graph Foundation Models (NeurIPS 2025 · similarity 0.86)
- Exploitation of a Latent Mechanism in Graph Contrastive Learning: Representation Scattering (NeurIPS 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
