Latent Semantic Space Transforming Techniques for Efficient Representation Learning and Applications

This thesis proposes three different latent semantic space transforming techniques that transfer one space into a new latent semantic space or map both two spaces into a common latent semantic space for learning efficient representations. These learned representations can contain useful features and deep hidden semantic information. Furthermore, efficient representations not only can improve the accuracy of prediction models in industrial communities and increase the productivity of human beings, but also be beneficial for different applications related to artificial intelligence, for example, smart manufacturing, smart recommendation, smart finance, and smart healthcare, etc.