Xiaoxiao Ma

Xiaoxiao Ma

I am a Senior Applied Scientist at Oracle with 9+ years of experience across industry and research, including work at Amazon and internships in Australia. My work and research focus on agentic AI systems, natural language processing, large language models, automatic prompt optimization, RAG, graph neural networks, recommender systems, anomaly detection, deep generative models, and deep learning.

Revisiting Graph-Level Anomaly Detection: From Partially to Fully Unsupervised Learning

Published in Proceedings of the ACM Web Conference 2026, 2026

This work revisits graph-level anomaly detection by moving from the common partially unsupervised setting, where training data is assumed to contain only normal graphs, to a fully unsupervised setting that learns directly from mixed real-world data. It introduces uncertainty-aware learning to narrow that gap and studies frameworks that can train without an expert filtering stage to remove anomalies beforehand. The goal is to make graph-level anomaly detection more practical for realistic deployment conditions.

Recommended citation:
Download Paper