Workshop Anomaly and Change Point Detection

The workshop will take place in Kaiserslautern. It is planned to organize a part of the event at Fraunhofer ITWM and a part at the premises of DFKI. More detailed information will follow as soon as the detailed program is available.
 

English
 

The topics of "Anomaly and Change Point Detection" are becoming more and more important. In the current development, more and more data is available, not least with regard to Germany's digitization initiative. This enables the use of Machine Learning (ML) methods and their further development in more and more areas.

At the same time, errors and changes are integrated in the data, which we would like to detect and use. Here, many research questions are still open. This workshop aims to provide a platform for exchange and discussion.

The research areas of "Anomaly and Change Point Detection" cover many different fields and are used in applications to answer a wide variety of questions. As an example, the following topics are mentioned here (other topics are conceivable):

• Anomaly Detection
• Change Point Detection
• Outlier Detection
• Unsupervised Machine Learning (ML)
• Data Analysis
• Mixed Type Datasets (e.g., tabular data with categorical and numerical columns)
• Dataset Comparison
• Dataset Similarity
• Transferability of Results to Internal Datasets
• Dataset Generation
• Data Synthesis
• Explainability
   

The topics are also becoming more and more important in various areas in practice. This is because more and more data is available due to current developments, not least with regard to Germany's digitalisation initiative. This enables the use of machine learning (ML) methods and their further development. The detection of anomalies or outliers always means discovering patterns in data that are not expected and cannot be classified in a defined normal state.

Well-known examples from the application and practice include:

• In production: In the manufacturing industry, anomaly detection is important for improving quality control, for example. A distinction is often made between one-off errors (anomaly) or general misadjustments of the machine (change point/structure break). Then, for example, the sensors can be adjusted.
• With climate change in general, outliers in the data occure frequently and the real change detectable in the general trend. However, an example from the past also shows exceptions. In the case of the ozone measurement in the 1980s: The ozone hole had already been discovered before, but was deleted from the data as an anomaly. The data was therefore misinterpreted.
• Parking data during the Corona pandemic: When predicting parking availability, major events are often the outliers, as here, for example, on a Sunday, there are suddenly hardly any spaces available, whereas at Corona the general parking situation was minimised, as fewer people were on the road at all.
• In the large area of financial data, anomaly detection is versatile. It can be used to identify fraud or manipulation in the network, but also to detect minor changes in the usage behaviour of customers and thus better predict their behaviour.