Deep Neural Networks and Data for Automated Driving
Robustness, Uncertainty Quantification, and Insights Towards Safety
Book
Paperback/Softback
440 pages
978-3-031-01234-1 (ISBN)
Description
This open access book brings together the latest developments from industry and research on automated driving and artificial intelligence.
Environment perception for highly automated driving heavily employs deep neural networks, facing many challenges. How much data do we need for training and testing? How to use synthetic data to save labeling costs for training? How do we increase robustness and decrease memory usage? For inevitably poor conditions: How do we know that the network is uncertain about its decisions? Can we understand a bit more about what actually happens inside neural networks? This leads to a very practical problem particularly for DNNs employed in automated driving: What are useful validation techniques and how about safety?
This book unites the views from both academia and industry, where computer vision and machine learning meet environment perception for highly automated driving. Naturally, aspects of data, robustness, uncertainty quantification, and, last but not least, safety are at the core of it. This book is unique: In its first part, an extended survey of all the relevant aspects is provided. The second part contains the detailed technical elaboration of the various questions mentioned above.
Environment perception for highly automated driving heavily employs deep neural networks, facing many challenges. How much data do we need for training and testing? How to use synthetic data to save labeling costs for training? How do we increase robustness and decrease memory usage? For inevitably poor conditions: How do we know that the network is uncertain about its decisions? Can we understand a bit more about what actually happens inside neural networks? This leads to a very practical problem particularly for DNNs employed in automated driving: What are useful validation techniques and how about safety?
This book unites the views from both academia and industry, where computer vision and machine learning meet environment perception for highly automated driving. Naturally, aspects of data, robustness, uncertainty quantification, and, last but not least, safety are at the core of it. This book is unique: In its first part, an extended survey of all the relevant aspects is provided. The second part contains the detailed technical elaboration of the various questions mentioned above.
More details
Edition
1st ed. 2022
Language
English
Place of publication
Cham
Switzerland
Target group
Professional and scholarly
Illustrations
X, 440 p.
Dimensions
Height: 235 mm
Width: 155 mm
ISBN-13
978-3-031-01234-1 (9783031012341)
Copyright in bibliographic data and cover images is held by Nielsen Book Services Limited or by the publishers or by their respective licensors: all rights reserved.
Schweitzer Classification
Person
Tim Fingscheidt received the Dipl.-Ing. degree in Electrical Engineering in 1993 and the Ph.D. degree in 1998 from RWTH Aachen University, Germany, both with distinction. He joined AT&T Labs, Florham Park, NJ, USA, for a PostDoc in 1998 and Siemens AG (Mobile Devices), Munich, Germany, in 1999, heading a signal processing development team. After a stay with Siemens Corporate Technology, Munich, Germany, from 2005 to 2006, he became Full Professor with the Institute for Communications Technology, Technische Universitaet (TU) Braunschweig, Germany, holding the Chair of "Signal Processing and Machine Learning". His research interests are machine learning in vision and time series such as speech, with focus on environment perception, signal classification, coding, and enhancement. He is founder of the TU Braunschweig Deep Learning Lab (tubs.DLL), a graduate student research thinks tank being active in publicly funded and industry research projects. Many of his projects have been dealing with automotive applications. Since 2018, he has been actively involved in the large-scale national research projects AI Platform Concept, AI Validation, AI Delta Learning, and AI Data Tooling, contributing research in robust semantic segmentation, monocular depth estimation, domain adaptation, corner case detection, and learned image coding. Prof. Fingscheidt received numerous national and international awards for his publications; among these, three CVPR workshop best paper awards in 2019, 2020, and 2021. He is interested in where academia meets industry and where machine learning meets highly automated driving.
Hanno Gottschalk studied Physics and Mathematics and received diploma degrees from the Ruhr University Bochum in 1995 and 1997, respectively. After finishing his Ph.D. on Mathematical Physics in 1999, he joined the University La Sapienza of Rome for a PostDoc year, before continuing his academic career as PostDoc at Bonn University, where he habilitated in mathematics in 2003. Since 2005, he was lecturer (C2) at the University of Bonn and joined Siemens Energy from 2007-2011 as a Core Competency Owner for probabilistic design. Since 2011, he is Professor for stochastics at the University of Wuppertal. In 2018, he became co-founding Director of the Interdisciplinary Center for Machine Learning and Data Analytics (IZMD) of the University of Wuppertal. His research in the field of deep learning is focused on uncertainty and safety for deep learning perception algorithms. Applications lie in the field of false positive and false negative prediction and detection and retrieval of out of distribution objects. Apart from bi-lateral work with Volkswagen and Aptiv, he is member of the AI Validation, AI Delta Learning, and AI Data Tooling consortia within the AI family of large-scale projects. Hanno Gottschalk brings his special knowledge as statistician and mathematician to the field of automated driving and combines this with cutting edge technology in deep learning.
Sebastian Houben studied Mathematics and Computer Science at the University in Hagen and graduated in 2009. He pursued Ph.D. studies at the Ruhr University of Bochum graduating with distinction in 2015. After his postdoctoral studies at the University of Bonn, he was appointed Junior Professor for Applied Computer Science at the Ruhr University of Bochum where he headed the Group of Real-time Computer Vision. As of early 2020, he is a senior researcher with the Fraunhofer Institute for Intelligent Analysis and Information Systems. His research interests cover computer vision and environment perception in autonomous robotics, in particular in the field of automated driving. Within the consortium KI-Absicherung and the competency center Machine-Learning-Rhein-Ruhr (ML2R), he represents the topic Trustworthy AI and is particularly interested in practical methods for explainability of black-box models, uncertainty estimation in neural networks, and visual analytics. Sebastian Houben believes that artificial intelligence would be an even stronger technology if it was simpler, more robust, and safer to use. His role at Fraunhofer allows him to accompany this transfer from the research laboratories into practical applications.
Hanno Gottschalk studied Physics and Mathematics and received diploma degrees from the Ruhr University Bochum in 1995 and 1997, respectively. After finishing his Ph.D. on Mathematical Physics in 1999, he joined the University La Sapienza of Rome for a PostDoc year, before continuing his academic career as PostDoc at Bonn University, where he habilitated in mathematics in 2003. Since 2005, he was lecturer (C2) at the University of Bonn and joined Siemens Energy from 2007-2011 as a Core Competency Owner for probabilistic design. Since 2011, he is Professor for stochastics at the University of Wuppertal. In 2018, he became co-founding Director of the Interdisciplinary Center for Machine Learning and Data Analytics (IZMD) of the University of Wuppertal. His research in the field of deep learning is focused on uncertainty and safety for deep learning perception algorithms. Applications lie in the field of false positive and false negative prediction and detection and retrieval of out of distribution objects. Apart from bi-lateral work with Volkswagen and Aptiv, he is member of the AI Validation, AI Delta Learning, and AI Data Tooling consortia within the AI family of large-scale projects. Hanno Gottschalk brings his special knowledge as statistician and mathematician to the field of automated driving and combines this with cutting edge technology in deep learning.
Sebastian Houben studied Mathematics and Computer Science at the University in Hagen and graduated in 2009. He pursued Ph.D. studies at the Ruhr University of Bochum graduating with distinction in 2015. After his postdoctoral studies at the University of Bonn, he was appointed Junior Professor for Applied Computer Science at the Ruhr University of Bochum where he headed the Group of Real-time Computer Vision. As of early 2020, he is a senior researcher with the Fraunhofer Institute for Intelligent Analysis and Information Systems. His research interests cover computer vision and environment perception in autonomous robotics, in particular in the field of automated driving. Within the consortium KI-Absicherung and the competency center Machine-Learning-Rhein-Ruhr (ML2R), he represents the topic Trustworthy AI and is particularly interested in practical methods for explainability of black-box models, uncertainty estimation in neural networks, and visual analytics. Sebastian Houben believes that artificial intelligence would be an even stronger technology if it was simpler, more robust, and safer to use. His role at Fraunhofer allows him to accompany this transfer from the research laboratories into practical applications.
Content
Chapter 1. Inspect, Understand, Overcome: A Survey of Practical Methods for AI Safety.- Chapter 2. Does Redundancy in AI Perception Systems Help to Test for Super-Human Automated Driving Performance?.- Chapter 3. Analysis and Comparison of Datasets by Leveraging Data Distributions in Latent Spaces.- Chapter 4. Optimized Data Synthesis for DNN Training and Validation by Sensor Artifact Simulation.- Chapter 5. Improved DNN Robustness by Multi-Task Training With an Auxiliary Self-Supervised Task.- Chapter 6. Improving Transferability of Generated Universal Adversarial Perturbations for Image Classi?cation and Segmentation.- Chapter 7. Invertible Neural Networks for Understanding Semantics of Invariances of CNN Representations.- Chapter 8. Con?dence Calibration for Object Detection and Segmentation.- Chapter 9. Uncertainty Quanti?cation for Object Detection: Output- and Gradient-based Approaches.- Chapter 10. Detecting and Learning the Unknown in Semantic Segmentation.- Chapter 11. Evaluating Mixture-of-Expert Architectures for Network Aggregation.- Chapter 12. Safety Assurance of Machine Learning for Perception Functions.- Chapter 13. A Variational Deep Synthesis Approach for Perception Validation.- Chapter 14. The Good and the Bad: Using Neuron Coverage as a DNN Validation Technique.- Chapter 15. Joint Optimization for DNN Model Compression and Corruption Robustness.