awesome anomaly detection

A curated list of awesome anomaly detection resources
By hoya012

deep-learning machine-learning awesome machinelearning anomaly-detection anomaly anomalydetection awesome-anomaly-detection awesomeanomalydetection

awesome anomaly detection

A curated list of awesome anomaly detection resources. Inspired by awesome-architecture-search and awesome-automl.

Last updated: 2020/12/16

What is anomaly detection?

Anomaly detection is a technique used to identify unusual patterns that do not conform to expected behavior, called outliers. Typically, this is treated as an unsupervised learning problem where the anomalous samples are not known a priori and it is assumed that the majority of the training dataset consists of “normal” data (here and elsewhere the term “normal” means not anomalous and is unrelated to the Gaussian distribution). [Lukas Ruff et al., 2018; Deep One-Class Classification]

In general, Anomaly detection is also called Novelty Detection or Outlier Detection, Forgery Detection and Out-of-distribution Detection.

Each term has slightly different meanings. Mostly, on the assumption that you do not have unusual data, this problem is especially called One Class Classification, One Class Segmentation.

and Novelty Detection and Outlier Detection have slightly different meanings. Figure below shows the differences of two terms.

Also, typically there are three types of target data. (time-series data, and image data, video data)
In time-series data, it is aimed to detect a abnormal sections.
In image, video data, it is aimed to classify abnormal images or to segment abnormal regions, for example, defect in some manufacturing data.

Survey Paper

Table of Contents

Time-series anomaly detection (need to survey more..)

Video-level anomaly detection

Image-level anomaly detection
One Class (Anomaly) Classification target

Out-of-Distribution(OOD) Detection target

Unsupervised Anomaly Segmentation target

Contact & Feedback

If you have any suggestions about papers, feel free to mail me :)
- e-mail
- blog
- pull request