Towards an Understanding of Wide, Deep Neural Networks