RangeAD: Fast On-Model Anomaly Detection

TL;DR

RangeAD introduces a novel on-model anomaly detection method that leverages neuron output ranges within the primary model, offering efficient and accurate detection especially in high-dimensional tasks.

Contribution

The paper proposes RangeAD, a new algorithm for on-model anomaly detection that uses neuron-wise output ranges, reducing inference costs and improving performance.

Findings

RangeAD outperforms existing methods on high-dimensional tasks.

It achieves lower inference costs compared to separate AD models.

RangeAD demonstrates practical efficiency in real-world scenarios.

Abstract

In practice, machine learning methods commonly require anomaly detection (AD) to filter inputs or detect distributional shifts. Typically, this is implemented by running a separate AD model alongside the primary model. However, this separation ignores the fact that the primary model already encodes substantial information about the target distribution. In this paper, we introduce On-Model AD, a setting for anomaly detection that explicitly leverages access to a related machine learning model. Within this setting, we propose RangeAD, an algorithm that utilizes neuron-wise output ranges derived from the primary model. RangeAD achieves superior performance even on high-dimensional tasks while incurring substantially lower inference costs. Our results demonstrate the potential of the On-Model AD setting as a practical framework for efficient anomaly detection.