Robustifying machine learning through weakening supervision / Julian Lienen ; 1. Reviewer: Prof. Dr. Eyke Hüllermeier, Institute of Informatics, Ludwig Maximilian University of Munich, 2. Reviewer: Dr. Sébastien Destercke, HDR, Centre national de la recherche scientifique (CNRS), Université de Technologie de Compiègne, 3. Reviewer: Prof. Dr. Ralph Ewerth, L3S Research Center, Leibniz University Hannover, Technische Informationsbibliothek (TIB) Hannover, Supervisor: Prof. Dr. Eyke Hüllermeier. Paderborn, 2023
Inhalt
- Titlepage
- Abstract
- Acknowledgement
- Contents
- 1 Introduction
- 2 Supervised Learning: Harnessing Precise Labels
- 3 Weakly-Supervised Learning: Generalizing Models and Data
- 3.1 Overview
- 3.2 Superset Learning
- 3.2.1 Average-Based Superset Learning
- 3.2.2 Identification-Based Superset Learning
- 3.2.3 Fuzzy Supersets
- 3.3 Semi-Supervised Learning
- 3.3.1 Common Assumptions
- 3.3.2 Unsupervised Preprocessing
- 3.3.3 Self-Labeling
- 3.3.4 Intrinsically Semi-Supervised Methods
- 3.4 Relative Comparisons as a Form of Weak Supervision
- 4 From Label Smoothing to Label Relaxation
- 5 Credal Self-Supervised Learning
- 6 Conformal Credal Self-Supervised Learning
- 7 Mitigating Label Noise through Data Ambiguation
- 8 Instance Weighting through Data Imprecisiation
- 9 Robust Regression for Monocular Depth Estimation
- 10 Monocular Depth Estimation via Listwise Ranking using the Plackett-Luce Model
- 11 Conclusion and Outlook
- Bibliography
- A Appendix to From Label Smoothing to Label Relaxation
- B Appendix to Credal Self-Supervised Learning
- C Appendix to Conformal Credal Self-Supervised Learning
- D Appendix to Mitigating Label Noise through Data Ambiguation
- E Appendix to Robust Regression for Monocular Depth Estimation
- F Appendix to Monocular Depth Estimation via Listwise Ranking using the Plackett-Luce Model
- List of Source Code Repositories
- List of Figures
- Colophon
- Declaration