New paper out

The paper MARINA Meets Matrix Stepsizes: Variance Reduced Distributed Non-convex Optimization is now on arXiv. In this work, we combine matrix stepsizes with MARINA type variance reduction for sketched gradient descent, the proposed algorithm, so called det-MARINA, is analyzed in the distributed non-convex setting . We establish theoretically and empirically, that det-MARINA outperforms both MARINA and the distributed det-CGD algorithms in terms of iteration and communication complexities.