DEMON SGD Optimizer for TensorFlow Keras See the paper titled "Decaying momentum helps neural network training" (https://arxiv.org/pdf/1910.04952.pdf) Tested uner TensorFlow 2.2