Keras implementation of the algorithm found in the paper: Backprop without Learning Rates Through Coin Betting by Francesco Orabona and Tatiana Tommasi https://arxiv.org/pdf/1705.07795.pdf
Tensorflow implementation here: /~https://github.com/bremen79/cocob
Example usage in the mnist_mlp python script
All credit goes to Francesco Orabona and Tatiana Tommasi, I simply spent a few minutes adapting their implementation to work with keras