Louise: A Meta-Interpretive Learner for Efficient Multi-clause Learning of Large Programs

We present Louise, a new Meta-Interpretive Learner that performs efficient multi-clause learning, implemented in Prolog. Louise is efficient enough to learn programs that are too large to be learned with the current state-of-the-art MIL system, Metagol. Louise learns by first constructing the most general program in the hypothesis space of a MIL problem and then reducing this "Top program" by Plotkin's program reduction algorithm. In this extended abstract we describe Louise's learning approach and experimentally demonstrate that Louise can learn programs that are too large to be learned by our implementation of Metagol, Thelma.