LVQ Treatment for Zero-Shot Learning

In image classification, there are no labeled training instances for some classes, which therefore called unseen classes or test classes. To classify these zero-shot learning (ZSL) was developed, typically attempts to learn a mapping from the (visual) feature space semantic in represented by list of semantically meaningful attributes. However, fact that this is learned without using affects performance ZSL, known as domain shift problem. study, we propose apply vector quantization (LVQ) algorithm once determined. First and foremost, allows us refine prototypes with respect mapping, reduces effects Secondly, LVQ increases margin 1-NN classifier used resulting better classification. Moreover, work, consider range algorithms, initial advanced variants, applied them number state-of-the-art ZSL methods, then obtained their extensions. The experiments based on five benchmark datasets showed LVQ-empowered extensions methods superior original counterparts almost all settings.