Research map / Graph learning / Graph self-supervised learning
Permutation-Invariant Variational Autoencoder for Graph-Level Representation Learning
Robin Winter, Frank Noe, Djork-Arne Clevert
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 PIGAE instead of GAE for graph reconstruction | Graph autoencoding · Erdos-Renyi, Barabasi-Albert, and Ego graphs | NLL: 20.5 ± 0.9 vs 186 ± 3; 27.2 ± 0.9 vs 199 ± 3; 23.4 ± 0.5 vs 191 ± 4 | Improved |
| use PIGAE instead of GAE for graph reconstruction | Graph autoencoding · Erdos-Renyi, Barabasi-Albert, and Ego graphs | ROC-AUC: 98.3 ± 0.1 vs 57.9 ± 0.1; 96.7 ± 0.2 vs 57.4 ± 0.1; 97.8 ± 0.3 vs 59.1 ± 0.1 | Improved |
| use PIGAE instead of Graphite | Graph reconstruction · Erdos-Renyi, Barabasi-Albert, and Ego graphs | NLL: 20.5 ± 0.9 vs 196 ± 1; 27.2 ± 0.9 vs 192 ± 2; 23.4 ± 0.5 vs 183 ± 1 | Improved |
| use PIGAE instead of GAE* | Graph reconstruction · Erdos-Renyi, Barabasi-Albert, and Ego graphs | NLL: 20.5 ± 0.9 vs 222 ± 8; 27.2 ± 0.9 vs 236 ± 15; 23.4 ± 0.5 vs 197 ± 2 | Improved |
| add topological distances as edge features (ablation) | PIGAE · Erdos-Renyi, Barabasi-Albert, and Ego graphs | NLL: 20.5 ± 0.9 -> 19.5 ± 0.8; 27.2 ± 0.9 -> 15.2 ± 0.8; 23.4 ± 0.5 -> 22.4 ± 0.5 | Improved |
| add topological distances as edge features (ablation) | PIGAE · Erdos-Renyi, Barabasi-Albert, and Ego graphs | ROC-AUC: 98.3 ± 0.1 -> 99.4 ± 0.1; 96.7 ± 0.2 -> 99.5 ± 0.1 ; 97.8 ± 0.3 -> 98.8 ± 0.3 | Improved |
| use PIGAE representations instead of averaged GAE node embeddings | SVM graph-class prediction · Synthetic graph class prediction (29 classes) | accuracy: 0.65 ± 0.01 -> 0.83 ± 0.01 | Improved |
| use PIGAE representations instead of InfoGraph representations | SVM graph-class prediction · Synthetic graph class prediction (29 classes) | accuracy: 0.75 ± 0.02 -> 0.83 ± 0.01 | Improved |
| use PIGAE representations instead of Shortest Path Kernel | SVM graph-class prediction · Synthetic graph class prediction (29 classes) | accuracy: 0.50 ± 0.02 -> 0.83 ± 0.01 | Improved |
| use PIGAE representations instead of Weisfeiler-Lehman Sub-tree Kernel | SVM graph-class prediction · Synthetic graph class prediction (29 classes) | accuracy: 0.73 ± 0.01 -> 0.83 ± 0.01 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Graphite — Graphite: Iterative Generative Modeling of Graphs (ICML 2019) · better on 3 of 3 shared results
- InfoGraph — InfoGraph: Unsupervised and Semi-supervised Graph-Level Representation Learning via Mutual Information Maximization (ICLR 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Unsupervised Graph Embedding via Adaptive Graph Learning (TPAMI 2022 · similarity 0.90)
- Learning Representations for Hierarchies with Minimal Support (NeurIPS 2024 · similarity 0.88)
- Subgraph Invariant Learning Towards Large-Scale Graph Node Classification (AAAI 2025 · similarity 0.88)
- Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning (ICLR 2025 · similarity 0.88)
- Equivalence is All: A Unified View for Self-supervised Graph Learning (ICML 2025 · similarity 0.88)
- GraphCroc: Cross-Correlation Autoencoder for Graph Structural Reconstruction (NeurIPS 2024 · similarity 0.87)
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.
