Research map / Graph learning / Graph self-supervised learning
Unsupervised Joint k-node Graph Representations with Compositional Energy-Based Models
Leonardo Cotta, Carlos H. C. Teixeira, Ananthram Swami, Bruno Ribeiro
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 |
|---|---|---|---|
| train MHM-GNN weights instead of random weights (ablation) | MHM-GNN · Hyperedge Detection k=3, DBLP | balanced accuracy: 0.491 -> 0.773 | Improved |
| train MHM-GNN weights instead of random weights (ablation) | MHM-GNN · Hyperedge Detection k=3, Cora | balanced accuracy: 0.498 -> 0.618 | Improved |
| train MHM-GNN weights instead of random weights (ablation) | MHM-GNN · Hyperedge Detection k=4, Pubmed | balanced accuracy: 0.510 -> 0.701 | Improved |
| train MHM-GNN weights instead of random weights (ablation) | MHM-GNN · Hyperedge Detection k=5, Citeseer | balanced accuracy: 0.453 -> 0.703 | Improved |
| train MHM-GNN weights instead of random weights (ablation) | MHM-GNN · DAG Leaf Counting k=3, Cora | balanced accuracy: 0.497 -> 0.593 | Improved |
| train MHM-GNN weights instead of random weights (ablation) | MHM-GNN · DAG Leaf Counting k=5, Pubmed | balanced accuracy: 0.244 -> 0.330 | Improved |
| train MHM-GNN weights instead of random weights (ablation) | MHM-GNN · Hyperedge Detection k=5, Rent the Runway | balanced accuracy: 0.924 -> 0.943 | No clear effect |
| use MHM-GNN joint k-node representation instead of GraphSAGE/DGI node representations | GraphSAGE, DGI · Hyperedge Detection k=3, Cora | balanced accuracy: 0.490 -> 0.618 | Improved |
| use MHM-GNN instead of raw features on a task where structure is irrelevant | Raw Features · Hyperedge Detection k=5, Rent the Runway | balanced accuracy: 0.944 -> 0.943 | No clear effect |
| use MHM-GNN instead of GraphSAGE-mean | GS-mean · DAG Leaf Counting k=4, Cora | balanced accuracy: 0.245 -> 0.452 | Improved |
| use MHM-GNN instead of DGI | DGI · Hyperedge Detection k=3, DBLP | balanced accuracy: 0.497 -> 0.773 | Improved |
| add node2vec/DeepWalk transductive embeddings with features | node2vec · Hyperedge Detection k=4, Cora | balanced accuracy: 0.537 -> 0.626 | No clear effect |
Closest papers by meaning
- Memory-Based Graph Networks (ICLR 2020 · similarity 0.86)
- Distance Encoding: Design Provably More Powerful Neural Networks for Graph Representation Learning (NeurIPS 2020 · similarity 0.86)
- Labeling Trick: A Theory of Using Graph Neural Networks for Multi-Node Representation Learning (NeurIPS 2021 · similarity 0.86)
- Graph Neural Networks Powered by Encoder Embedding for Improved Node Learning (TPAMI 2026 · similarity 0.86)
- How Powerful are Graph Neural Networks? (ICLR 2019 · similarity 0.86)
- Exploring Consistency in Graph Representations: from Graph Kernels to Graph Neural Networks (NeurIPS 2024 · similarity 0.85)
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.
