Learning like humans with Deep Symbolic Networks

Abstract

Knowledge and Data together form the foundation of most AI systems - Symbolic or Semantic knowledge forms concepts and relations describing structures in a logical and human-interpretable way, allowing queries, reasoning, and inference. Sub-symbolic or Syntactic Data is little or non-structured sensor data, such as images or audio, that has high volume and is harder to interpret or program against by humans in their raw form. While there are often connections between them, they both have their own techniques for learning and adapting models, and this is done mostly separately today. Considering recent advances in deep learning, researchers are now reviewing existing and developing new methods for hybrid learning. where knowledge and data are used in conjunction to train inter-linked models that offer both the predictive strength and efficiency of data-based models, as well as the structure and transparency of knowledge-based models.