I’ve implemented SVRG (Stochastic Variance Reduction Gradient Descent) optimizer for Keras. The goal is to make this optimizer available in Keras as well, which may be beneficial in the case of RL as some papers claimed it is advantageous over Adam.

The paper: https://papers.nips.cc/paper/4937-accelerating-stochastic-gradient-descent-using-predictive-variance-reduction.pdfLink to the project: https://github.com/tilkb/SVRGoptimizerKeras