Structure learning for relational logistic regression: an ensemble approach

We consider the problem of learning Relational Logistic Regression (RLR). Unlike standard logistic regression, features RLR are first-order formulae with associated weight vectors instead scalar weights. turn to these vector-weighted and develop a algorithm based on recently successful functional-gradient boosting methods for probabilistic logic models. derive functional gradients show how weights can be learned simultaneously in an efficient manner. Our empirical evaluation data sets demonstrates superiority our approach over other RLR.