Injecting Domain Knowledge in Neural Networks: a Controlled Experiment on a Constrained Problem

Given enough data, Deep Neural Networks (DNNs) are capable of learning\ncomplex input-output relations with high accuracy. In several domains, however,\ndata is scarce or expensive to retrieve, while a substantial amount of expert\nknowledge is available. It seems reasonable that if we can inject this\nadditional information in the DNN, we could ease the learning process. One such\ncase is that of Constraint Problems, for which declarative approaches exists\nand pure ML solutions have obtained mixed success. Using a classical\nconstrained problem as a case study, we perform controlled experiments to probe\nthe impact of progressively adding domain and empirical knowledge in the DNN.\nOur results are very encouraging, showing that (at least in our setup)\nembedding domain knowledge at training time can have a considerable effect and\nthat a small amount of empirical knowledge is sufficient to obtain practically\nuseful results.\n