The representation and processing of uncertain knowledge is one of the major topics in AI and still a challenge. Different perspectives and priorities have led to numerous approaches to the field. KReate is a research project that aims at developing a common methodology for learning, modelling and inference in a relational probabilistic framework. In this common methodolgy, uncertain rules can be extracted from relational data, represented as probabilistic conditionals and used for inference. Therefore, first-order conditionals will serve as the principal knowledge representation concept to join these different fields, making an optimal exploration and usage of relational information possible by fine-tuning the involved processes. We will pursue both theoretical and experimental objectives in this project. In particular, we expect the approach to be developed to provide a powerful conditional-logical framework that will prove useful in practice. Moreover, a platform for learning, modelling and inference will be implemented to support quick and efficient programming as well as testing and practical applications.
DFG – Deutsche Forschungsgemeinschaft (grants BE 1700/7-1 and KE 1413/2-1)
For now, please see the main publications page.