Anomaly detection (AD), separating anomalies from normal data, has many applications across domains, from security to healthcare. While most previous works were shown to be effective for cases with fully or partially labeled data, that setting is in practice less common due to labeling being particularly tedious for this task. In this paper, we focus on fully unsupervised AD, in which the entire training dataset, containing both normal and anomalous samples, is unlabeled. To tackle this problem effectively, we propose to improve the robustness of one-class classification trained on self-supervised representations using a data refinement process. Our proposed data refinement approach is based on an ensemble of one-class classifiers (OCCs), each of which is trained on a disjoint subset of training data. Representations learned by self-supervised learning on the refined data are iteratively updated as the data refinement improves. We demonstrate our method on various unsupervised AD tasks with image and tabular data. With a 10% anomaly ratio on CIFAR-10 image data / 2.5% anomaly ratio on Thyroid tabular data, the proposed method outperforms the state-of-the-art one-class classifier by 6.3 AUC and 12.5 average precision / 22.9 F1-score.
翻译:异常检测(AD)将异常数据与正常数据区分开来,在从安全到医疗保健等多个领域都有许多应用。虽然大多数先前的工程都显示对带有全部或部分标签数据的案件有效,但由于标签特别繁琐,因此设置实际上不那么常见。在本文中,我们侧重于完全不受监督的AD, 包括正常和异常样本的整个培训数据集都不受监管,其中含有正常和异常样本,没有标签。为有效解决这一问题,我们提议通过数据改进过程,提高经过自我监督演示培训的单级分类的稳健性。我们建议的数据改进方法基于一个单级分类器(OCCs)的组合,每个分类器都受过不连续的培训。在改进数据的过程中,通过自我监督学习精炼数据而获得的演示内容被反复更新。我们用图像和表格数据来展示了我们各种不超超强的AD任务的方法。CRFAR-10图像数据中的10%异常率/Thyroid 表格数据的2.5%异常率。我们提出的方法是按ASroid ASRA-S-ARIA AS-A ASI ASI ASAL ASI ASU ASI ASI ASI ASIS 和 ASI ASIM ASI ASI ASI ASI ASI ASI ASI ASI ASI ASI ASI ASI ASI 平均 ASIS AS ASIS ASIS 和 ASI ASI ASIS ASI ASI ASI ASI ASI ASIS ASI ASI ASI ASI ASI ASY ASI AS AS AS AS AS ASIS ASI AS AS AS ASI 和 ASIS AS ASI ASIS AS ASIS AS AS AS AS AS ASI ASI ASI ASI ASI ASI AS AS AS ASIS AS AS AS AS AS AS AS AS AS AS AS AS AS AS AS AS AS AS AS AS ASIS AS AS AS AS AS AS AS