Evolutionary Router Feature Generation for Zero-Shot Graph Anomaly Detection with Mixture-of-Experts
This new MoE framework could revolutionize how we find hidden threats in complex networks.
Researchers have introduced EvoFG, a novel Mixture-of-Experts framework with evolutionary router feature generation for zero-shot graph anomaly detection. It tackles key routing challenges caused by distribution shifts across different graphs. The system uses an LLM-based generator and Shapley-guided evaluation to iteratively construct informative structural features. Extensive experiments on six benchmarks show EvoFG consistently outperforms state-of-the-art baselines, achieving strong and stable zero-shot performance for identifying anomalies.
Why It Matters
It enables more accurate detection of fraud, cyberattacks, and failures in complex systems without needing labeled data for each new graph.