Data-driven Detection of Anomalous Inputs End-to-end Neural Approaches to Detect Anomalies and Adversarial Examples
Year: 2023
Type: dissertation
Abstract: Anomalies are inputs that deviate from the given notion of normal. Anomaly detection methods automatically reveal such data points. In our research, we explore two main paths to combine data-driven learning methods like neural networks and anomaly detection: 1) the generation of meaningful synthetic anomalies and 2) the data-driven representation of inputs. Our research led to four anomaly detection methods, which are readily applicable to several important use cases of anomaly detection.
Authors: Schulze, Jan-Philipp, Eckert, Claudia 1965-, Technische Universität München
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6980866723