Research map / Graph learning / Graph self-supervised learning
Anomaly as Non-Conformity via Training-Free Graph Laplacian Energy Minimization
Jungwook Seo, Minjeong Kim, Younkwan Lee, Seungho Shin, Sungyong Baik
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 |
|---|---|---|---|
| add bipartite query-reference graph instead of non-bipartite graph (ablation) | k-NN + graph energy · MVTec-AD (1-shot) | image AUROC: 93.7 -> 95.8 | Improved |
| add bipartite query-reference graph instead of non-bipartite graph (ablation) | k-NN + graph energy · VisA (1-shot) | image AUROC: 86.3 -> 90.5 | Improved |
| add anchor-driven retrieval instead of plain k-NN retrieval (ablation) | ANoCo · MVTec-AD (1-shot) | image AUROC: 95.8 -> 97.9 | Improved |
| add anchor-driven retrieval instead of plain k-NN retrieval (ablation) | ANoCo · VisA (1-shot) | image AUROC: 90.5 -> 92.7 | Improved |
| add graph energy on a non-bipartite graph to k-NN (Mahalanobis) (ablation) | k-NN · VisA (1-shot) | image AUROC: 77.0 -> 86.3 | Improved |
| replace energy minimization with message passing (1 round) (ablation) | ANoCo · MVTec-AD (1-shot) | image AUROC: 87.7 vs 97.9 | Improved |
| use more message-passing rounds (1 to 2) (ablation) | message passing · MVTec-AD (1-shot) | image AUROC: 87.7 -> 82.7 | Hurt |
| use product of Euclidean energy and cosine dissimilarity as score instead of L2 only (ablation) | ANoCo · MVTec-AD (1-shot) | image AUROC: 97.6 -> 97.9 | No clear effect |
| use product score instead of cosine distance only (ablation) | ANoCo · VisA (1-shot) | image AUROC: 92.4 -> 92.7 | No clear effect |
| sweep query stabilization coefficient Lambda_q from 1e-24 to 100 (ablation) | ANoCo · MVTec-AD and VisA (1-shot) | AUROC | No clear effect |
| increase reference shots from 1 to 4 | ANoCo · Real-IAD | image AUROC: 65.4 -> 83.8 | Improved |
| use full-shot references instead of 1-shot | ANoCo · VisA | image AUROC: 92.7 -> 98.7 | Improved |
Closest papers by meaning
- CurvGAD: Leveraging Curvature for Enhanced Graph Anomaly Detection (ICML 2025 · similarity 0.86)
- A Unified Latent Schrodinger Bridge Diffusion Model for Unsupervised Anomaly Detection and Localization (CVPR 2025 · similarity 0.86)
- One-to-Normal: Anomaly Personalization for Few-shot Anomaly Detection (NeurIPS 2024 · similarity 0.86)
- Correcting Deviations from Normality: A Reformulated Diffusion Model for Multi-Class Unsupervised Anomaly Detection (CVPR 2025 · similarity 0.85)
- UniGAD: Unifying Multi-level Graph Anomaly Detection (NeurIPS 2024 · similarity 0.85)
- Rethinking Reconstruction-based Graph-Level Anomaly Detection: Limitations and a Simple Remedy (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.
