Preference Neural Network

This paper proposes a novel label ranker network to learn the relationship between labels to solve ranking and classification problems. The Preference Neural Network (<italic>PNN</italic>) uses <italic>spearman</italic> correlation gradient ascent and two new activation functions, positive smooth staircase (<italic>PSS</italic>), and smooth staircase (<italic>SS</italic>) that accelerate the ranking by creating almost deterministic preference values. <italic>PNN</italic> is proposed in two forms, fully connected simple Three layers and Preference Net (<italic>PN</italic>), where the latter is the deep ranking form of <italic>PNN</italic> to learning feature selection using ranking to solve images classification problem. <italic>PN</italic> uses a new type of ranker kernel to generate a feature map. <italic>PNN</italic> outperforms five previously proposed methods for label ranking, obtaining state-of-the-art results on label ranking, and <italic>PN</italic> achieves promising results on <italic>CFAR-100</italic> with high computational efficiency.