3 个月前

深度无监督异常检测

深度无监督异常检测

摘要

本文提出了一种在完全无监督设置下检测大规模数据集异常的新方法。该算法的核心思想是学习正常数据背后的潜在表示。为此,我们采用最新发展的适用于高维数据处理的聚类技术,从而为正常数据的选取提供一个可靠的初始依据。在此基础上,我们基于聚类结果与表示学习,迭代地构建正常候选数据子集,并训练自编码器模型。利用所学自编码器的重构误差作为评分函数,评估数据的正常程度。在多个公开基准数据集上的实验结果表明,所提出的方法在性能上优于当前最先进的无监督异常检测技术,在大多数情况下其表现可与半监督方法相媲美。

基准测试

基准方法指标
unsupervised-anomaly-detection-with-specifiedDeep Unsup.
AUC-ROC: 0.866
unsupervised-anomaly-detection-with-specified-1Deep Unsup.
AUC-ROC: 0.740
unsupervised-anomaly-detection-with-specified-10Deep Unsup.
AUC-ROC: 0.525
unsupervised-anomaly-detection-with-specified-11Deep Unsup.
AUC-ROC: 0.765
unsupervised-anomaly-detection-with-specified-12Deep Unsup.
AUC-ROC: 0.545
unsupervised-anomaly-detection-with-specified-13Deep Unsup.
AUC-ROC: 0.779
unsupervised-anomaly-detection-with-specified-14Deep Unsup.
AUC-ROC: 0.884
unsupervised-anomaly-detection-with-specified-15Deep Unsup.
AUC-ROC: 0.956
unsupervised-anomaly-detection-with-specified-16Deep Unsup.
AUC-ROC: 0.847
unsupervised-anomaly-detection-with-specified-17Deep Unsup.
AUC-ROC: 0.891
unsupervised-anomaly-detection-with-specified-18Deep Unsup.
AUC-ROC: 0.868
unsupervised-anomaly-detection-with-specified-19Deep Unsup.
AUC-ROC: 0.862
unsupervised-anomaly-detection-with-specified-20Deep Unsup.
AUC-ROC: 0.906
unsupervised-anomaly-detection-with-specified-21Deep Unsup.
AUC-ROC: 0.847
unsupervised-anomaly-detection-with-specified-22Deep Unsup.
AUC-ROC: 0.847
unsupervised-anomaly-detection-with-specified-23Deep Unsup.
AUC-ROC: 0.835
unsupervised-anomaly-detection-with-specified-24Deep Unsup.
AUC-ROC: 0.773
unsupervised-anomaly-detection-with-specified-25Deep Unsup.
AUC-ROC: 0.856
unsupervised-anomaly-detection-with-specified-26Deep Unsup.
AUC-ROC: 0.862
unsupervised-anomaly-detection-with-specified-27Deep Unsup.
AUC-ROC: 0.878
unsupervised-anomaly-detection-with-specified-5Deep Unsup.
AUC-ROC: 0.869
unsupervised-anomaly-detection-with-specified-6Deep Unsup.
AUC-ROC: 0.702
unsupervised-anomaly-detection-with-specified-7Deep Unsup.
AUC-ROC: 0.689
unsupervised-anomaly-detection-with-specified-8Deep Unsup.
AUC-ROC: 0.384
unsupervised-anomaly-detection-with-specified-9Deep Unsup.
AUC-ROC: 0.841

用 AI 构建 AI

从想法到上线——通过免费 AI 协同编程、开箱即用的环境和市场最优价格的 GPU 加速您的 AI 开发

AI 协同编程
即用型 GPU
最优价格
立即开始

Hyper Newsletters

订阅我们的最新资讯
我们会在北京时间 每周一的上午九点 向您的邮箱投递本周内的最新更新
邮件发送服务由 MailChimp 提供
深度无监督异常检测 | 论文 | HyperAI超神经