# From Zero to Hero: Cold-Start Anomaly Detection Source: [From Zero to Hero: Cold-Start Anomaly Detection]() Collection index: [llms.txt]() Data updated: 2026-01-02T06:05:11.801Z Authors: Tal Reiss, George Kour, Naama Zwerdling, Ateret Anaby-Tavor, Yedid Hoshen Published: 2024-05-30T17:59:51.000Z ## 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. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Out-of-scope Intent Detection on CLINC-Banking (C-Bank) r=2.5% contamination (test)](): Out-of-scope Intent Detection; CLINC-Banking (C-Bank) r=2.5% contamination (test); AUC (10% Contamination): 87. - [Out-of-scope intent detection on CLINC-CreditCards OOS (test)](): Out-of-scope Intent Detection; CLINC-CreditCards OOS (test); AUC (10% Threshold): 86.4. - [Out-of-scope Intent Detection on CLINC-Cards (C-Cards) r=2.5% contamination (test)](): Out-of-scope Intent Detection; CLINC-Cards (C-Cards) r=2.5% contamination (test); AUC2 (10%): 85.2. - [Out-of-Scope Query Detection on Banking77-OOS (test)](): Out-of-Scope Query Detection; Banking77-OOS (test); AUC^2 (10%): 83.3. - [Out-of-Scope Query Detection on CLINC-Banking (C-Bank) (test)](): Out-of-Scope Query Detection; CLINC-Banking (C-Bank) (test); AUC (10% Threshold): 84.4. - [Out-of-scope Intent Detection on Banking77 (r=2.5% contamination, test)](): Out-of-scope Intent Detection; Banking77 (B77) r=2.5% contamination (test); AUC2 (10% Subset): 0.806. ## Original sources and implementation [Original paper]() [PDF]() [Implementation]()