PAC-Wrap: Semi-Supervised PAC Anomaly Detection

TL;DR

PAC-Wrap is a versatile framework that provides rigorous Probably Approximately Correct guarantees for semi-supervised anomaly detection methods, enhancing safety-critical applications like autonomous driving.

Contribution

It introduces a novel wrapping approach that endows existing anomaly detectors with provable error bounds in a semi-supervised setting.

Findings

PAC-Wrap effectively provides error guarantees across various datasets.

It can be applied to multiple existing anomaly detection algorithms.

The method enhances safety-critical applications with rigorous bounds.

Abstract

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.