No Multiplication? No Floating Point? No Problem! Training Networks for Efficient Inference