Title: Neuro-Symbolic Class Expression Learning
Abstract: We consider the problem of learning class expressions (also known as concepts) using knowledge graphs with description logics semantics as background knowledge. Using this form of explainable machine learning on real data has long been considered impractical due to the significant runtimes it required. However, their increase usage in real applications and on the Web make them first-class citizens of modern information systems and hence requires time-efficient machine learning solutions. In the talk, we present some recent advances based on neuro-symbolic that expedite class expression learning by several orders of magnitude while maintaining varying degrees of explainability.