| Abstract: |  This thesis deals with a design of novel classification model from the field of relational machine learning, which uses a structure inspired by artificial neural networks for representation of logic program. The logic program syntax is a subset of Datalog but extended with disjunctions with weight coefficients. The weights are used for learning the model with modified backpropagation method known from artificial neural networks. Then the work presents experimental implementation of the proposed model and validates its accuracy. The accuracy is close to the best algorithms from the literature. Moreover, the presented system can be further extended in various directions. So far, it implements disjunction weights learning but not structure learning, which could potentially boost its performance significantly. This remains a challenge for future work.
|
|---|