Could you explain the role of adjustment in anomaly detection?

[ZhoumingLiu]

What is the role of adjustment in anomaly detection, and why is this step necessary?

I noticed that adjustment is also applied in another project, 'Anomaly Transformer', and some people are puzzled by this step. #issue

If the author sees this question, could you please explain it? How should I apply it in the industrial field?

[tianzhou2011]

I believe this concept was proposed before the anomaly transformer. Essentially, it means that if I identify one anomalous point in a sequence, it infers that I have detected all anomalies within that sequence. This setting could be particularly useful in the AIOps industry, where the focus is often on recall. The goal is to alert human operators to potential issues without the critical need for pinpoint accuracy, as the details can be examined later by human. Thus, this approach is well-suited for some real-world industrial applications, making the task significantly easier. However, its applicability depends on the specific context; in scientific settings where precise identification of each individual point is essential, it may not be suitable.

On Tue, Oct 8, 2024 at 3:36 PM ZhoumingLiu @.***> wrote:

0b872ee5679735beff1b1b699e2384f.png (view on web) https://github.com/user-attachments/assets/b92a7f3f-9a31-42fb-9f8a-63a28659079c What is the role of adjustment in anomaly detection, and why is this step necessary?

I noticed that adjustment is also applied in another project, 'Anomaly Transformer', and some people are puzzled by this step. #issue https://github.com/thuml/Anomaly-Transformer/issues/65

If the author sees this question, could you please explain it? How should I apply it in the industrial field?

— Reply to this email directly, view it on GitHub https://github.com/DAMO-DI-ML/NeurIPS2023-One-Fits-All/issues/45, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB3JGOYNJ6JBWA7EDG7YJLLZ2ODQ3AVCNFSM6AAAAABPRRIRIOVHI2DSMVQWIX3LMV43ASLTON2WKOZSGU3TEMZVGIYTGOI . You are receiving this because you are subscribed to this thread.Message ID: @.***>

[ZhoumingLiu]

我相信这个概念是在异常转换器出现之前提出的。本质上，这意味着如果我识别出序列中的一个异常点，则推断我已经检测到该序列中的所有异常。这种设置在 AIOps 行业中可能特别有用，因为该行业的重点往往是召回。目标是提醒人类操作员注意潜在问题，而不需要精确定位，因为细节可以由人类稍后检查。因此，这种方法非常适合一些现实世界的工业应用，使任务变得容易得多。然而，它的适用性取决于具体情况；在科学环境中，精确识别每个单独的点至关重要，它可能并不合适。 … ZhoumingLiu @.> 于 2024 年 10 月 8 日星期二下午 3:36 写道： 0b872ee5679735beff1b1b699e2384f.png (在网页上查看) < https://github.com/user-attachments/assets/b92a7f3f-9a31-42fb-9f8a-63a28659079c > 调整在异常检测中的作用是什么，为什么需要这一步？我注意到调整也应用于另一个项目 'Anomaly Transformer'，有些人对这一步感到困惑。#issue < thuml/Anomaly-Transformer#65 > 如果作者看到这个问题，你能解释一下吗？我应该如何将它应用于工业领域？ — 直接回复此电子邮件，在 GitHub 上查看 < #45 >，或取消订阅 < https://github.com/notifications/unsubscribe-auth/AB3JGOYNJ6JBWA7EDG7YJLLZ2ODQ3AVCNFSM6AAAAABPRRIRIOVHI2DSMVQWIX3LMV43ASLTON2WKOZSGU3TEMZVGIYTGOI >。您收到此邮件是因为您订阅了此主题。消息 ID：@.>

Thank you for your response, it has been very helpful for my understanding.