Time: 14:00 - 14:30
FREE AND OPEN FOR ALL TO ATTEND
We present a new approach for extracting rules, expressed in Horn logic, from neural network models. The approach is based on the exact learning model, in which a learner interacts with a teacher (the neural network model) via queries and counterexamples in order to learn an abstract target concept. We employ Angluin’s algorithm for learning Horn theories with membership and equivalence queries and evaluate the approach empirically. Our experimental results for extracting Horn theories use a dataset in the domain of ophthalmic optics and a synthetic one.