By using a technique for extracting various abstract internal symbolic knowledge from a deep learning model through inductive learning, this research aims to convert abstract internal symbolic knowledge into a human-readable knowledge graph, and to develop symbolic (KBN) reflecting the degree of uncertainty about internal symbolic knowledge. For knowledge expansion, the research also aims to develop a technology that applies the method of learning rules between similar knowledge existing in the established knowledge base (KBH), reflecting certainty, to new knowledge.
Technology to convert internal expressions into human-readable formats
Knowledge expression technology that expresses the knowledge extracted from the neural network model trained on noisy data differently depending on the degree of uncertainty
Technology to expand the knowledge base by deriving new symbolic knowledge through inductive learning
Development of an extraction engine and interface that can integrate multimodal knowledge from neural network models (visual, auditory, language)