Deep neural network initialization with decision trees

In this work a novel, automated process for determining an appropriate deep neural network architecture and weight initialization based on decision trees is presented. The method maps a collection of decision trees trained on the data into a collection of initialized neural networks, with the structure of the network determined by the structure of the tree. These models, referred to as "deep jointly-informed neural networks", demonstrate high predictive performance for a variety of datasets. Furthermore, the algorithm is readily cast into a Bayesian framework, resulting in accurate and scalable models that provide quantified uncertainties on predictions.