Few-Shot Graph Anomaly Detection via Dual-Level Knowledge Distillation
Graph anomaly detection is crucial in many high-impact applications across diverse fields. In anomaly detection tasks, collecting plenty of annotated data tends to be costly and laborious. As a result, few-shot learning has been explored to address the issue by requiring only a few labeled samples t...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-01-01
|
| Series: | Entropy |
| Subjects: | |
| Online Access: | https://www.mdpi.com/1099-4300/27/1/28 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832588561191272448 |
|---|---|
| author | Xuan Li Dejie Cheng Luheng Zhang Chengfang Zhang Ziliang Feng |
| author_facet | Xuan Li Dejie Cheng Luheng Zhang Chengfang Zhang Ziliang Feng |
| author_sort | Xuan Li |
| collection | DOAJ |
| description | Graph anomaly detection is crucial in many high-impact applications across diverse fields. In anomaly detection tasks, collecting plenty of annotated data tends to be costly and laborious. As a result, few-shot learning has been explored to address the issue by requiring only a few labeled samples to achieve good performance. However, conventional few-shot models may not fully exploit the information within auxiliary sets, leading to suboptimal performance. To tackle these limitations, we propose a dual-level knowledge distillation-based approach for graph anomaly detection, DualKD, which leverages two distinct distillation losses to improve generalization capabilities. In our approach, we initially train a teacher model to generate prediction distributions as soft labels, capturing the entropy of uncertainty in the data. These soft labels are then employed to construct the corresponding loss for training a student model, which can capture more detailed node features. In addition, we introduce two representation distillation losses—short and long representation distillation—to effectively transfer knowledge from the auxiliary set to the target set. Comprehensive experiments conducted on four datasets verify that DualKD remarkably outperforms the advanced baselines, highlighting its effectiveness in enhancing identification performance. |
| format | Article |
| id | doaj-art-7f8a34f1533440fdabf83151c4153c24 |
| institution | Kabale University |
| issn | 1099-4300 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Entropy |
| spelling | doaj-art-7f8a34f1533440fdabf83151c4153c242025-01-24T13:31:43ZengMDPI AGEntropy1099-43002025-01-012712810.3390/e27010028Few-Shot Graph Anomaly Detection via Dual-Level Knowledge DistillationXuan Li0Dejie Cheng1Luheng Zhang2Chengfang Zhang3Ziliang Feng4National Key Laboratory of Fundamental Science on Synthetic Vision, Sichuan University, Chengdu 610065, ChinaNational Key Laboratory of Fundamental Science on Synthetic Vision, Sichuan University, Chengdu 610065, ChinaNational Key Laboratory of Fundamental Science on Synthetic Vision, Sichuan University, Chengdu 610065, ChinaNational Key Laboratory of Fundamental Science on Synthetic Vision, Sichuan University, Chengdu 610065, ChinaNational Key Laboratory of Fundamental Science on Synthetic Vision, Sichuan University, Chengdu 610065, ChinaGraph anomaly detection is crucial in many high-impact applications across diverse fields. In anomaly detection tasks, collecting plenty of annotated data tends to be costly and laborious. As a result, few-shot learning has been explored to address the issue by requiring only a few labeled samples to achieve good performance. However, conventional few-shot models may not fully exploit the information within auxiliary sets, leading to suboptimal performance. To tackle these limitations, we propose a dual-level knowledge distillation-based approach for graph anomaly detection, DualKD, which leverages two distinct distillation losses to improve generalization capabilities. In our approach, we initially train a teacher model to generate prediction distributions as soft labels, capturing the entropy of uncertainty in the data. These soft labels are then employed to construct the corresponding loss for training a student model, which can capture more detailed node features. In addition, we introduce two representation distillation losses—short and long representation distillation—to effectively transfer knowledge from the auxiliary set to the target set. Comprehensive experiments conducted on four datasets verify that DualKD remarkably outperforms the advanced baselines, highlighting its effectiveness in enhancing identification performance.https://www.mdpi.com/1099-4300/27/1/28anomaly detectiongraph neural networkcross entropyknowledge distillation |
| spellingShingle | Xuan Li Dejie Cheng Luheng Zhang Chengfang Zhang Ziliang Feng Few-Shot Graph Anomaly Detection via Dual-Level Knowledge Distillation Entropy anomaly detection graph neural network cross entropy knowledge distillation |
| title | Few-Shot Graph Anomaly Detection via Dual-Level Knowledge Distillation |
| title_full | Few-Shot Graph Anomaly Detection via Dual-Level Knowledge Distillation |
| title_fullStr | Few-Shot Graph Anomaly Detection via Dual-Level Knowledge Distillation |
| title_full_unstemmed | Few-Shot Graph Anomaly Detection via Dual-Level Knowledge Distillation |
| title_short | Few-Shot Graph Anomaly Detection via Dual-Level Knowledge Distillation |
| title_sort | few shot graph anomaly detection via dual level knowledge distillation |
| topic | anomaly detection graph neural network cross entropy knowledge distillation |
| url | https://www.mdpi.com/1099-4300/27/1/28 |
| work_keys_str_mv | AT xuanli fewshotgraphanomalydetectionviaduallevelknowledgedistillation AT dejiecheng fewshotgraphanomalydetectionviaduallevelknowledgedistillation AT luhengzhang fewshotgraphanomalydetectionviaduallevelknowledgedistillation AT chengfangzhang fewshotgraphanomalydetectionviaduallevelknowledgedistillation AT ziliangfeng fewshotgraphanomalydetectionviaduallevelknowledgedistillation |