Using ML autoencoders for anomaly detection in accelerator controls

Machine learning is a promising new tool for anomaly detection in particle accelerators, but an inherent imbalance of normal and anomalous data, plus a lack of labeled data, makes using traditional techniques like supervised learning unrealistic. Enter, autoencoders. These neural networks form a compressed representation and reconstruction of the input data, which are very useful for situations like these. In this webinar, Senior Research Scientist Jon Edelen will explain autoencoders and use them for two common anomaly detection problems: dimensionality reduction and reconstruction analysis. He will give an overview of a few real-world examples and wrap up with a worked example in a Jupyter notebook, which will be made available after the webinar.

Видео Using ML autoencoders for anomaly detection in accelerator controls канала RadiaSoft