Research map / Label-efficient and robust learning / Knowledge distillation
Remembering Normality: Memory-guided Knowledge Distillation for Unsupervised Anomaly Detection
Zhihao Gu, Liang Liu, Xu Chen, Ran Yi, Jiangning Zhang, Yabiao Wang, Chengjie Wang, Annan Shu, Guannan Jiang, Lizhuang Ma
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 NR Memory with orthogonal loss to RD (ablation) | RD baseline · MVTec AD | Image AUROC: 98.4 -> 98.7 | Improved |
| add normality memorization loss L_NM to NR Memory (ablation) | RD with NR Memory · MVTec AD | Image AUROC: 98.7 -> 99.1 | Improved |
| combine NR Memory, L_Orth and L_NM (ablation) | RD baseline · MVTec AD | Image AUROC: 98.4 -> 99.6 | Improved |
| increase memory items from 0 to 50 per stage (ablation) | MemKD · MVTec AD | Image AUROC: 98.9 -> 99.6 | Improved |
| increase memory items from 50 to 100 per stage (ablation) | MemKD · MVTec AD | Image AUROC: 99.6 -> 99.4 | Hurt |
| apply NR Memory to forward distillation (MKD) (ablation) | MKD · MVTec AD | Image AUROC: 97.5 -> 98.7 | Improved |
| use NR Memory with normality embedding learning instead of MemAE memory (ablation) | RD baseline · MPDD | Pixel PRO: 86.2 -> 94.9 | Improved |
| NR Memory adds storage and time over RD | RD baseline · MPDD | memory / inference time: 0.0M -> 0.3M; 0.31s -> 0.33s | Hurt |
| MemKD versus PatchCore on pixel AUROC | MemKD · VisA | Pixel AUROC: 98.8 -> 98.4 | Hurt |
Closest papers by meaning
- Unlocking the Potential of Reverse Distillation for Anomaly Detection (AAAI 2025 · similarity 0.89)
- Co-Progression Knowledge Distillation with Knowledge Prototype for Industrial Anomaly Detection (AAAI 2025 · similarity 0.86)
- Normal-Abnormal Guided Generalist Anomaly Detection (NeurIPS 2025 · similarity 0.86)
- CNC: Cross-modal Normality Constraint for Unsupervised Multi-class Anomaly Detection (AAAI 2025 · similarity 0.86)
- Unsupervised Continual Anomaly Detection with Contrastively-Learned Prompt (AAAI 2024 · similarity 0.85)
- SetAD: Semi-Supervised Anomaly Learning in Contextual Sets (NeurIPS 2026 · 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.
