Review for NeurIPS paper: Follow the Perturbed Leader: Optimism and Fast Parallel Algorithms for Smooth Minimax Games

The paper provides a follow the perturbed leader algorithm and analysis that can obtain better regret bounds when loss/gradient sequence is predictable. The proofs relies on using the equivalent regularization view of FTPL. The authors also provide an application of this result to providing a parallelizable algorithms for solving smooth convex concave saddlepoint games Most of the reviewers found the result interesting. Please address the concerns of the reviewer. Personally, I find the predictable sequences result interesting.