While the popularity of statistical, probabilistic and exhaustive machine learning techniques still increases, relational and logic approaches are still a niche market in research. While the former approaches focus on predictive accuracy, the latter ones prove to be indispensable in knowledge discovery.
In this paper we present a relational description of machine learning problems. We demonstrate how common ensemble learning methods as used in classifier learning can be reformulated in a relational setting. It is shown that multimodal logics and relational data analysis with rough sets are closely related. Finally, we give an interpretation of logic programs as approximations of hypotheses.
It is demonstrated that at a certain level of abstraction all these methods unify into one and the same formalisation which nicely connects to multimodal operators.
Bitte loggen Sie sich ein, um Zugang zu diesem Inhalt zu erhalten