Efficiency testing of openset learning methods in image classification

Kovács, László and Palencsár, Enikő and Bán, Péter (2025) Efficiency testing of openset learning methods in image classification. In: Proceedings of the International Conference on Formal Methods and Foundations of Artificial Intelligence. Eszterházy Károly Katolikus Egyetem Líceum Kiadó, Eger, pp. 129-139. ISBN 9789634963035

Preview

Text fmfai2025_pp129-139.pdf - Published Version Download (642kB) | Preview

Abstract

The problem of detecting untrained categories may cause efficiency degradation in many application areas, because the real-word domains are usually dynamic, or the available data set may be incomplete. Despite the relatively high cost of related misclassification errors, the field of openset learning is an underinvestigated domain in machine learning. The main goal of this paper is to investigate the efficiency of current technologies for the openset learning problem on a standard benchmark image dataset. As the results of the performed comparison tests show that the widely proposed standard methods do not provide good results, in many cases the hybrid methods can dominate the usual approaches. In the paper, we present a novel extended threshold method that provides better accuracies than the usual benchmark methods.

Actions (login required)

Edit Item