arXiv:2511.04086v1 Announce Type: new 
Abstract: With the rapid growth of graph-structured data in critical domains, unsupervised graph-level anomaly detection (UGAD) has become a pivotal task. UGAD seeks to identify entire graphs that deviate from normal behavioral patterns. However, most Graph Neural Network (GNN) approaches implicitly assume that the training set is clean, containing only normal graphs, which is rarely true in practice. Even modest contamination by anomalous graphs can distort learned representations and sharply degrade performance. To address this challenge, we propose DeNoise, a robust UGAD framework explicitly designed for contaminated training data. It jointly optimizes a graph-level encoder, an attribute decoder, and a structure decoder via an adversarial objective to learn noise-resistant embeddings. Further, DeNoise introduces an encoder anchor-alignment denoising mechanism that fuses high-information node embeddings from normal graphs into all graph embeddings, improving representation quality while suppressing anomaly interference. A contrastive learning component then compacts normal graph embeddings and repels anomalous ones in the latent space. Extensive experiments on eight real-world datasets demonstrate that DeNoise consistently learns reliable graph-level representations under varying noise intensities and significantly outperforms state-of-the-art UGAD baselines.

تعتبر المقدمة الحديثة لـ DeNoise تقدمًا كبيرًا في الكشف عن الشذوذ على مستوى الرسوم البيانية غير الخاضعة للإشراف (UGAD)، وهو مجال حاسم مع استمرار نمو البيانات الهيكلية في الرسوم البيانية. تتناول هذه الطريقة تحديًا شائعًا في هذا المجال: الافتراض بأن مجموعات التدريب خالية من الشذوذ، وهو ما لا يكون صحيحًا في كثير من الأحيان. من خلال تحديد الرسوم البيانية الكاملة التي تنحرف عن الأنماط الطبيعية، تعزز DeNoise موثوقية الكشف عن الشذوذ في تطبيقات متنوعة، مما يجعلها أداة قيمة للباحثين والممارسين على حد سواء.

La reciente introducción de DeNoise representa un avance significativo en la detección de anomalías a nivel de grafos no supervisada (UGAD), un área crucial a medida que los datos estructurados en grafos continúan expandiéndose en importancia. Este método aborda un desafío común en el campo: la suposición de que los conjuntos de entrenamiento están libres de anomalías, lo cual no es el caso en muchas ocasiones. Al identificar de manera efectiva grafos enteros que se desvían de los patrones normales, DeNoise mejora la fiabilidad de la detección de anomalías en diversas aplicaciones, convirtiéndose en una herramienta valiosa para investigadores y profesionales.

La récente introduction de DeNoise marque une avancée significative dans la détection d'anomalies au niveau des graphes non supervisée (UGAD), un domaine crucial alors que les données structurées en graphes continuent de gagner en importance. Cette méthode aborde un défi courant dans le domaine : l'hypothèse selon laquelle les ensembles de données d'entraînement sont exempts d'anomalies, ce qui n'est souvent pas le cas. En identifiant efficacement des graphes entiers qui s'écartent des modèles normaux, DeNoise améliore la fiabilité de la détection d'anomalies dans diverses applications, en faisant un outil précieux pour les chercheurs et les praticiens.

The recent introduction of DeNoise marks a significant advancement in unsupervised graph-level anomaly detection (UGAD), a crucial area as graph-structured data continues to expand in importance. This method addresses a common challenge in the field: the assumption that training datasets are free from anomalies, which is often not the case. By effectively identifying entire graphs that deviate from normal patterns, DeNoise enhances the reliability of anomaly detection in various applications, making it a valuable tool for researchers and practitioners alike.

DeNoise: Learning Robust Graph Representations for Unsupervised Graph-Level Anomaly Detection

One More Thing in AI – Your Shortcut to AI Mastery

DeNoise: Learning Robust Graph Representations for Unsupervised Graph-Level Anomaly Detection

Was this article worth reading? Share it

One More Thing in AI

LucidQuery AI

Airparser

Humanize AI

Graphite Note

ZeroGPT.org

Ready to build your own newsroom?