Vendi Anomaly Scores for Efficient and Accurate Anomaly Detection
Removing a data point reveals how much it contributes to dataset diversity—and provides a score for spotting anomalies.
The Vendi Anomaly Score captures local redundancy and global data structure without density estimation or partitioning heuristics. The authors say it requires no hyperparameter tuning and scales linearly with dataset size. They report state-of-the-art performance across 148 benchmark embedding datasets drawn from 10 source datasets and four embedding architectures. Large-scale ImageNet experiments also showed robustness as contamination rates and dataset size increased.
ArXiv · AI/CL/LG's note
The Vendi Anomaly Score captures local redundancy and global data structure without density estimation or partitioning heuristics. The authors say it requires no hyperparameter tuning and scales linearly with dataset size. They report state-of-the-art performance across 148 benchmark embedding datasets drawn from 10 source datasets and four embedding architectures. Large-scale ImageNet experiments also showed robustness as contamination rates and dataset size increased.
ArXiv · AI/CL/LG's note
score 4