In recent years, neural systems have displayed highly effective learning ability and superior perception intelligence, but have been found to lack cognitive ability with effective reasoning. In the contrast, symbolic systems have exceptional cognitive intelligence, but their learning capabilities are poor compared to neural systems. Considering the advantages and disadvantages of both methodologies, an ideal solution is to combine neural systems and symbolic systems, an approach that produces neural-symbolic systems with powerful perception and cognition. In this paper, we survey recent advances in neural-symbolic systems from four perspectives: challenges, methods, applications, and future directions. This paper aims to advance this emerging area of research by providing researchers with a holistic and comprehensive overview of the field that highlights the state-of-the-art and identifies promising future research directions.
View on arXiv