Research map / Graph learning / Graph self-supervised learning
OrphicX: A Causality-Inspired Latent Variable Model for Interpreting Graph Neural Networks
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 latent causal-factor information flow with the full OrphicX objective | GNN explanation · BA-SHAPES and TREE-CYCLES | explanation accuracy | Helped |
| Use latent causal-factor information flow with the full OrphicX objective | GNN explanation · Mutag and NCI1 | explanation accuracy | Helped |
| Add the fidelity penalty (ablation) | OrphicX · Mutag | explanation accuracy: 0.829 -> 0.854 | Helped |
| Add the fidelity penalty (ablation) | OrphicX · NCI1 | explanation accuracy: 0.633 -> 0.832 | Helped |
| Add the sparsity penalty (ablation) | OrphicX · Mutag | explanation accuracy: 0.804 -> 0.854 | Helped |
| Add the sparsity penalty (ablation) | OrphicX · NCI1 | explanation accuracy: 0.824 -> 0.832 | Helped |
| Add the VGAE ELBO term (ablation) | OrphicX · Mutag | explanation accuracy: 0.594 -> 0.854 | Helped |
| Add the VGAE ELBO term (ablation) | OrphicX · NCI1 | explanation accuracy: 0.633 -> 0.832 | Helped |
| Add noise to causal latent factors | Pre-trained GNN on Mutag · Mutag | prediction accuracy: 0.935 -> 0.826 | Hurt |
| Add noise to spurious latent factors | Pre-trained GNN on Mutag · Mutag | prediction accuracy: 0.935 -> 0.926 | No clear effect |
| Use deconfounding under node-attribute distribution shift | OrphicX · Mutag | explanation accuracy: 77.2 -> 67.1; 78.8 -> 71.5; 83.2 -> 81.5 | Mixed |
| Use deconfounding under node-attribute distribution shift | OrphicX · NCI1 | explanation accuracy: 77.1 -> 71.6; 81.3 -> 79.2; 85.4 -> 87.3 | Mixed |
Closest papers by meaning
- Generative Causal Explanations for Graph Neural Networks (ICML 2021 · similarity 0.92)
- Rethinking Causal Relationships Learning in Graph Neural Networks (AAAI 2024 · similarity 0.89)
- A Recipe for Causal Graph Regression: Confounding Effects Revisited (ICML 2025 · similarity 0.88)
- Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks (ICLR 2024 · similarity 0.88)
- Beyond Topological Self-Explainable GNNs: A Formal Explainability Perspective (ICML 2025 · similarity 0.87)
- Factor Graph-based Interpretable Neural Networks (ICLR 2025 · 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.