Keywords:AI, Machine learning, Concept drift
In the operation of AI systems using machine learning technologies, a change in data distribution, called “concept drift”, is one of the main causes to degrade their performance. However, it is not only difficult to predict those changes before starting operation but also a quite costly task to assign their true labels by hand. In this paper, we present the results of a survey on the detection and adaptation of concept drift from the well-known methods published so far to their latest ones. We especially introduce the methods without using true labels of unlabeled operational data (or using only a limited number of them) in detection and adaptation.
Authentication for paper PDF access
A password is required to view paper PDFs. If you are a registered participant, please log on the site from Participant Log In.
You could view the PDF with entering the PDF viewing password bellow.