Machine Learning in Safety-Critical Applications
Description
Designed as a jumping-off point for engineers and decision-makers, this book provides a broad view of machine learning safety. It gives a validation road-map for developers and users of safety-critical systems where human lives are at stake. It addresses the limitations of machine learning systems at every stage of their lifecycle, and provides an overview of techniques to mitigate risks.
Chapters are structured to ease understanding of concepts: For each stage in the machine learning lifecycle, the reader is given a brief overview, list of terms, example methods, application methods, and resources for further investigation.
Specifically addresses safety-critical systems where human users are at risk, physical property can be damaged, or financial losses are possible
Covers datasets, robustness, reliability, interpretability, explainability, verification, validation, and operational safety
Covers relevant safety standards
Easy-to-digest overview of critical and fast-moving field
Extensive bibliography
Those working in safety-critical areas, such as autonomous transportation, medical diagnostics and treatments, robotics and manufacturing, and financial systems, will find this book valuable.
More details
Persons
Dr.-Ing. Oliver De Candido earned his doctoral degree ( summa cum laude ) from the Technical University of Munich (TUM) in 2023, specializing in Machine Learning (ML) safety for automated driving systems. He has led research teams developing ML safety evaluation frameworks, contributed to international safety standards for ML in road vehicles, and mentored dozens of students and researchers through university programs and specialized ML safety courses. With over 20 peer-reviewed publications, he bridges research with safety-critical applications, bringing both scientific rigor and hands-on expertise to validating ML systems.
Dr.-Ing. Michael Koller holds a doctorate ( summa cum laude ) from the Technical University of Munich (TUM), where he specialized in Machine Learning (ML) for communication systems. His research spans more than 25 peer-reviewed publications across ML applications in communications engineering, signal processing, and automated driving. It was through his industry work that he identified a timely need for a practical guide to ML validation.
Content
Chapter 1. Introduction.- Chapter 2. Interpretability and Explainability.- Chapter 3. Robustness and Reliability.- 4. Data Coverage.- Chapter 5. Verification and Validation.- Chapter 6. Monitoring and Operation.- Chapter 7. Summary and Outlook.- 8. Further Reading.