Anomalous Anomaly Detection

Muyeed Ahmed, Iulian Neamtiu
{"title":"Anomalous Anomaly Detection","authors":"Muyeed Ahmed, Iulian Neamtiu","doi":"10.1109/AITest55621.2022.00009","DOIUrl":null,"url":null,"abstract":"Anomaly Detection (AD) is an integral part of AI, with applications ranging widely from health to finance, manufacturing, and computer security. Though AD is popular and various AD algorithm implementations are found in popular toolkits, no attempt has been made to test the reliability of these implementations. More generally, AD verification and validation are lacking. To address this need, we introduce an approach and study on 4 popular AD algorithms as implemented in 3 popular tools, as follows. First, we checked whether implementations can perform their basic task of finding anomalies in datasets with known anomalies. Next, we checked two basic properties, determinism and consistency. Finally, we quantified differences in algorithms’ outcome so users can get a idea of variations that can be expected when using different algorithms on the same dataset. We ran our suite of analyses on 73 datasets that contain anomalies. We found that, for certain implementations, validation can fail on 10–73% of datasets. Our analysis has revealed that five implementations suffer from nondeterminism (19–98% of runs are nondeterministic), and 10 out of 12 implementation pairs are inconsistent.","PeriodicalId":427386,"journal":{"name":"2022 IEEE International Conference On Artificial Intelligence Testing (AITest)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2022-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2022 IEEE International Conference On Artificial Intelligence Testing (AITest)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/AITest55621.2022.00009","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1

Abstract

Anomaly Detection (AD) is an integral part of AI, with applications ranging widely from health to finance, manufacturing, and computer security. Though AD is popular and various AD algorithm implementations are found in popular toolkits, no attempt has been made to test the reliability of these implementations. More generally, AD verification and validation are lacking. To address this need, we introduce an approach and study on 4 popular AD algorithms as implemented in 3 popular tools, as follows. First, we checked whether implementations can perform their basic task of finding anomalies in datasets with known anomalies. Next, we checked two basic properties, determinism and consistency. Finally, we quantified differences in algorithms’ outcome so users can get a idea of variations that can be expected when using different algorithms on the same dataset. We ran our suite of analyses on 73 datasets that contain anomalies. We found that, for certain implementations, validation can fail on 10–73% of datasets. Our analysis has revealed that five implementations suffer from nondeterminism (19–98% of runs are nondeterministic), and 10 out of 12 implementation pairs are inconsistent.
异常检测
异常检测(AD)是人工智能的一个组成部分,应用范围广泛,从健康到金融、制造和计算机安全。尽管AD很流行,并且在流行的工具包中可以找到各种AD算法实现,但没有人尝试测试这些实现的可靠性。更普遍的是,缺乏AD验证和确认。为了满足这一需求,我们介绍了一种方法,并研究了在3种流行工具中实现的4种流行AD算法,如下所示。首先,我们检查了实现是否可以执行在已知异常的数据集中发现异常的基本任务。接下来,我们检查了两个基本属性,决定论和一致性。最后,我们量化了算法结果的差异,这样用户就可以了解在同一数据集上使用不同算法时可能出现的变化。我们在73个包含异常的数据集上运行了我们的分析套件。我们发现,对于某些实现,在10-73%的数据集上验证可能失败。我们的分析显示,有5个实现存在不确定性(19-98%的运行是不确定性的),12个实现对中有10个不一致。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
自引率
0.00%
发文量
0
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信