Entscheiden, was gemessen werden soll: Methoden zur Wichtigkeit und Auswahl von Informationen
Abstract
Active feature acquisition (AFA) addresses the problem of learning which features to acquire when acquisition is costly. As a prerequisite, feature importance under missing data is studied to identify which features are worth acquiring. The challenge of evaluating AFA policies from retrospective data is then formalized, introducing assumptions and estimators for reliable offline evaluation under distribution shift.
Fields of Study & Subjects
Machine Learning and Algorithms
Advanced Bandit Algorithms Research
Reinforcement Learning in Robotics
Publication details
- Journal / source
- mediaTUM (Technical University of Munich)
- Publication date
- 15 January 2026
- Language
- EN
- Document type
- Thesis