Zeitschriftenartikel:
A. Mozelli, N. TaheriNejad, A. Jantsch:
"A Study on Confidence: an Unsupervised Multi-Agent Machine Learning Experiment";
Ieee Design & Test,
39
(2021),
3;
8 S.
Kurzfassung englisch:
Computational Self-Awareness (CSA) is a growing field that has been applied to various applications, which often uses Machine Learning (ML). One of the key metrics for assessing the quality of both CSA and ML systems is confidence, which has been used in many applications recently. Confidence has shown a great promise in improving systems´ performance, in particular regarding the reliability of operations. However, from an engineering point of view, the nature of confidence as a metric has been an open question. Understanding the nature of confidence can help the better usage of the concept and, consequently, the design of better systems. Uncovering the true nature of confidence, however, is not currently within our reach. Therefore, in this work, we take one step in that direction by designing a socially-inspired experiment to investigate the nature of confidence in the context of (self-)learning. Our experiment shows that among the two candidates discussed in the literature, probability is a better metric for confidence. This observation sheds light on this open question and marks an entry point for further investigating the concept of confidence as a metric in ML and CSA.
"Offizielle" elektronische Version der Publikation (entsprechend ihrem Digital Object Identifier - DOI)
http://dx.doi.org/10.1109/MDAT.2021.3078341
Elektronische Version der Publikation:
https://publik.tuwien.ac.at/files/publik_300504.pdf
Erstellt aus der Publikationsdatenbank der Technischen Universität Wien.