Year: 2023
Type: dissertation
Abstract: In this dissertation, we address the challenge of validating Machine Learning (ML)-based Highly Automated Driving (HAD) functions. We develop validation safety arguments considering various aspects of ML algorithms: From the embeddings learned by different architectures, through the dataset distributions used to train them, to an interpretable design of an ML algorithm. The presented methods can be used individually or complementarily to validate various aspects of ML-based HAD functions.
Authors: De Candido, Oliver Thomas, Wolfgang Utschick, Technische Universität München
Language: English
Subfield: Safety, Risk, Reliability and Quality
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7025440043