Committee: Allen Malony (chair), Daniel Lowd, Michel Kinsy
Area Exam(Sep 2015)
Keywords: Deep Learning, Semantic Data Mining, Semantic Deep Learning, Ontology, Formal Knowledge, Domain Knowledge
Artificial intelligence and machine learning research is dedicated to building intelligent artifacts that can imitate or even transcend the cognitive abilities of human beings. To emulate human cognitive abilities with intelligent artifacts, one must first render machines capable of capturing critical aspects of sensory data, with adequate data representations and performing reasoning and inference with formal knowledge representations. In recent years, the research in deep learning and knowledge engineering has made wide impact on the two problems of data and knowledge representations. Deep learning is a set of machine learning algorithms that attempt to model data representations through many layers of non-linear transformations. Hierarchical, distributed, and efficient data representations can be learned through deep learning models with proper training algorithms. The learned data representation can disentangle the hidden explanation factors and variations in the input data that are critical for further artificial intelligence and machine learning tasks. Additionally, the research in knowledge engineering has frequently focused on modeling the high level human cognitive abilities, such as reasoning, making inferences, and validation. The formal knowledge representation facilitates knowledge reusing and sharing in a machine processable way. It also promotes many advances in the field of semantic data mining which refers to the data mining tasks that systematically incorporate domain knowledge, especially formal semantics, into the data mining process. Empirical studies have attested that formal knowledge representations can make positive influences in all stages of both the data mining and machine learning processes. Inspired by the success of both deep learning and semantic data mining, we hypothesize that formal knowledge representations have the potential to assist in the deep learning process as well. In this report, we summarize the advances in both deep learning and semantic data mining in recent years. We illustrate how learning models with deeper architectures are capable of constructing better data representations for further artificial intelligence and machine learning tasks. We also demonstrate how formal knowledge representation can assist in data mining process at all data mining stages, from various perspectives. At last, we present our thoughts and intuitions on semantic deep learning, which addresses the topic of learning deep data representation with the assistance of formal knowledge representation.