Get Started
Home
Topics
Search
Library
Research questionHow can graph-level anomaly detection transfer to new domains with little or no target data?Graph-level anomaly detectors may learn patterns tied to their training domain, making anomalies harder to identify after deployment in a different domain. The difficulty is greater when the target domain provides no training data or only a few normal examples.
AI
Machine Learning
Research Paper
Statistical Machine Learning
Technology
Latest papersRecent research connected to this question, newest first.GLASS: Graph-Language Alignment with Spherical Scoring for Transferable Graph-Level Anomaly DetectionThe source presents GLASS, which aligns structure-aware graph representations with instruction-aware text embeddings and encodes graph properties in Graph Descriptor Prompts. It supports zero-shot detection without target-domain training data and few-shot calibration with a small normal reference set; evidence comes from twelve benchmarks across three meta-domains, using AUROC and rank comparisons with graph-level anomaly-detection baselines.research paper · Sep 4, 2026
Related questions
How can zero-shot graph anomaly detectors remain reliable on unlabeled targets under domain shift?How can we detect multivariate IoT traffic anomalies across deployments without labels or raw-feature sharing?How can dynamic-graph anomaly detection capture short-term interaction behavior in event streams?How can we detect anomalous higher-order associations in hypergraphs without labeled examples?