Deep Anomaly Detection by Residual Adaptation
This work addresses the challenge of detecting anomalies in complex datasets for applications like security or quality control, representing a strong specific gain rather than a foundational advancement.
The paper tackles the problem of deep anomaly detection in high-dimensional data by proposing a method that augments pretrained networks with residual corrections, resulting in a parameter-efficient approach that improves disentanglement and achieves state-of-the-art performance, raising mean AUC from 96.1 to 99.0 on the CIFAR-10 benchmark.
Deep anomaly detection is a difficult task since, in high dimensions, it is hard to completely characterize a notion of "differentness" when given only examples of normality. In this paper we propose a novel approach to deep anomaly detection based on augmenting large pretrained networks with residual corrections that adjusts them to the task of anomaly detection. Our method gives rise to a highly parameter-efficient learning mechanism, enhances disentanglement of representations in the pretrained model, and outperforms all existing anomaly detection methods including other baselines utilizing pretrained networks. On the CIFAR-10 one-versus-rest benchmark, for example, our technique raises the state of the art from 96.1 to 99.0 mean AUC.