Explainable Artificial Intelligence in Anomaly Detection

Current artificial intelligences are incredibly powerful tools that learn from large amounts of data. In many cases, however, they are also high complex, which makes it difficult to understand their behavior. Explainable artificial intelligence aims to provide simple explanations for decisions from complex models. This work focuses on explanations in anomaly detection, a specialized domain of artificial intelligence where small numbers of anomalies must be found in largely normal data.

This work can focus either on evaluation of existing anomaly detectors, or the construction of specialized models that are easy to explain.

Supervisor: Julian Tritscher