Setting language bias automatically for relational learner

Relational learning algorithms tend to take a really long time to find the target query due to large search space. Conventionally an expert writes a guided hyperparameter setting also called as language bias to reduce this search space. In this work, we present our hyperparameter framework, Automode which generates the language bias automatically. Our ILP learner, Castor takes this language bias as an input to find the target query. We have observed that running Castor with automatically generated language bias usually has a longer running time. However, it removes any need for an expert to guide the search. Additionally, we also show how the generalization part of a learning algorithm can be improved to use this language bias.

Major Advisor: Arash Termehchy
Committee: Prasad Tadepalli
Committee: Alan Fern