Mastersthesis,
Simulative Evaluation of (In-)Confident Machine Learning for User-based Active Learning
University of Würzburg, (December 2023)
Abstract
Machine learning is becoming more and more prevalent in networking, e.g, network monitoring and managament. Though, expert knowledge is always valuable, sometimes irreplaceable. As the machine learning model may not always be confident, one idea is to relay such inconfident decisions to an expert, i.e., the admin, which can decide what to do with this decision and potentially relabel it, called user-based active learning. The goal of this thesis is to perform a simulative analysis regarding the influence of various parameters, such as the confidence threshold (i.e., when to relay the information) or the processing time and qualification level of the admin (i.e., a fast but not always correct expert, slow but accurate etc.). For more information, read the linked presentation slide and/or message katharina.dietz@uni-wuerzburg.de.
