Akademik Veri Yönetim Sistemi
2nd International Congress on Human-Computer Interaction, Optimization and Robotic Applications (HORA), Türkiye, 26 - 27 Haziran 2020, ss.510-513
With deep learning achieving more successful results than traditional machine learning methods, researches in the field of computer vision have evolved towards this area. However, in order to obtain successful models in deep learning methods, it needs a large number of training samples similar to traditional machine learning methods. In order to meet this requirement, auxiliary information of visual data has been used in recent years. Zero-shot learning methods focused on the compatibility functions of image embeddings and class embeddings, and researches aimed at better representation of class embeddings on visual data. In this paper, recent studies on zero-shot learning have been examined and evaluated.