Projected gradient methods for nonconvex and stochastic optimization: new complexities and auto-conditioned stepsizes

Open in new window