DomiKnowS: A Library for Integration of Symbolic Domain Knowledge in Deep Learning

We demonstrate a library for the integration of domain knowledge in deep\nlearning architectures. Using this library, the structure of the data is\nexpressed symbolically via graph declarations and the logical constraints over\noutputs or latent variables can be seamlessly added to the deep models. The\ndomain knowledge can be defined explicitly, which improves the models'\nexplainability in addition to the performance and generalizability in the\nlow-data regime. Several approaches for such an integration of symbolic and\nsub-symbolic models have been introduced; however, there is no library to\nfacilitate the programming for such an integration in a generic way while\nvarious underlying algorithms can be used. Our library aims to simplify\nprogramming for such an integration in both training and inference phases while\nseparating the knowledge representation from learning algorithms. We showcase\nvarious NLP benchmark tasks and beyond. The framework is publicly available at\nGithub(https://github.com/HLR/DomiKnowS).\n