RangeAD: Fast On-Model Anomaly Detection
This work addresses the need for efficient anomaly detection in machine learning applications, offering a practical framework that reduces computational overhead while maintaining high performance.
The paper tackles the problem of anomaly detection by proposing On-Model AD, a setting that leverages information from a related primary model instead of using a separate AD model, and introduces RangeAD, an algorithm based on neuron-wise output ranges that achieves superior performance with lower inference costs.
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.