Ontology-Based Deep Restricted Boltzmann Machine

Deep neural networks are known for their capabilities for automatic feature learning from data. For this reason, previous research has tended to interpret deep learning techniques as data-driven methods, while few advances have been made from knowledge-driven perspectives. We propose to design a semantic rich deep learning model from a knowledge driven perspective, by introducing formal semantics into deep learning process. We propose ontology-based deep restricted Boltzmann machine (OB-DRBM), in which we use ontology to guide architecture design of deep restricted Boltzmann machines (DRBM), as well as to assist in their training and validation processes. Our model learns a set of related semantic-rich data representations from both formal semantics and data distribution. Representations in this set correspond to concepts at various semantic levels in a domain ontology. We show that our model leads to an improved performance, when compared with conventional deep learning models in classification tasks.