PAC-Wrap: Semi-Supervised PAC Anomaly Detection
This addresses the need for reliable anomaly detection in safety-critical domains like autonomous driving, offering a novel approach to enhance existing methods with rigorous guarantees.
The paper tackles the problem of providing provable error bounds for anomaly detection in safety-critical applications by introducing PAC-Wrap, a method that wraps around existing anomaly detectors to offer Probably Approximately Correct guarantees on false negative and false positive rates, with experiments showing broad effectiveness across various detectors and datasets.
Anomaly detection is essential for preventing hazardous outcomes for safety-critical applications like autonomous driving. Given their safety-criticality, these applications benefit from provable bounds on various errors in anomaly detection. To achieve this goal in the semi-supervised setting, we propose to provide Probably Approximately Correct (PAC) guarantees on the false negative and false positive detection rates for anomaly detection algorithms. Our method (PAC-Wrap) can wrap around virtually any existing semi-supervised and unsupervised anomaly detection method, endowing it with rigorous guarantees. Our experiments with various anomaly detectors and datasets indicate that PAC-Wrap is broadly effective.