Knowledge Infused Learning (K-IL): Towards Deep Incorporation of Knowledge in Deep Learning

Learning the underlying patterns in data goes beyond instance-based\ngeneralization to external knowledge represented in structured graphs or\nnetworks. Deep learning that primarily constitutes neural computing stream in\nAI has shown significant advances in probabilistically learning latent patterns\nusing a multi-layered network of computational nodes (i.e., neurons/hidden\nunits). Structured knowledge that underlies symbolic computing approaches and\noften supports reasoning, has also seen significant growth in recent years, in\nthe form of broad-based (e.g., DBPedia, Yago) and domain, industry or\napplication specific knowledge graphs. A common substrate with careful\nintegration of the two will raise opportunities to develop neuro-symbolic\nlearning approaches for AI, where conceptual and probabilistic representations\nare combined. As the incorporation of external knowledge will aid in\nsupervising the learning of features for the model, deep infusion of\nrepresentational knowledge from knowledge graphs within hidden layers will\nfurther enhance the learning process. Although much work remains, we believe\nthat knowledge graphs will play an increasing role in developing hybrid\nneuro-symbolic intelligent systems (bottom-up deep learning with top-down\nsymbolic computing) as well as in building explainable AI systems for which\nknowledge graphs will provide scaffolding for punctuating neural computing. In\nthis position paper, we describe our motivation for such a neuro-symbolic\napproach and framework that combines knowledge graph and neural networks.\n