From Zero to Hero: Cold-Start Anomaly Detection

TL;DR

This paper introduces ColdFusion, a novel method for effectively adapting zero-shot anomaly detectors to contaminated observations in cold-start scenarios, supported by a new evaluation suite.

Contribution

It proposes ColdFusion, a method that combines zero-shot guidance with limited contaminated data for improved anomaly detection in cold-start settings.

Findings

ColdFusion outperforms existing methods in cold-start anomaly detection.

The evaluation suite enables standardized assessment of zero-shot anomaly detection methods.

ColdFusion effectively utilizes small contaminated datasets to enhance detection accuracy.

Abstract

When first deploying an anomaly detection system, e.g., to detect out-of-scope queries in chatbots, there are no observed data, making data-driven approaches ineffective. Zero-shot anomaly detection methods offer a solution to such "cold-start" cases, but unfortunately they are often not accurate enough. This paper studies the realistic but underexplored cold-start setting where an anomaly detection model is initialized using zero-shot guidance, but subsequently receives a small number of contaminated observations (namely, that may include anomalies). The goal is to make efficient use of both the zero-shot guidance and the observations. We propose ColdFusion, a method that effectively adapts the zero-shot anomaly detector to contaminated observations. To support future development of this new setting, we propose an evaluation suite consisting of evaluation protocols and metrics.