Experimented with and compared DFW neural network optimizer with SGD and ADAM on both vision and language tasks