RAIN (Recursive Anchoing IterRation)

December 10, 2024 · View on GitHub

Official codes for "Near-optimal algorithms for making the gradient small in stochastic minimax optimization."

We conduct numerical experiments on the following three hard instances:

  • a convex-concave bilinear function (in the folder src/bilinear_func).
  • a convex-concave (δ,ν)(\delta,\nu) function with additive Guassian noise (in the folder src/delta_func).
  • a nonconvex-nonconcave function under the comonotonicity condition (in the folder src/rho_func).

To reimplement the expeiments in our paper, please run

exp_gnorm.m

in the corresponding folders.