Randomized Subspace Nesterov Accelerated Gradient

ArXi:2605.00740v1 Announce Type: cross Randomized-subspace methods reduce the cost of first-order optimization by using only low-dimensional projected-gradient information, a feature that is attractive in forward-mode automatic differentiation and communication-limited settings. While Nestero acceleration is well understood for full-gradient and coordinate-based methods, obtaining accelerated methods for general subspace sketches that use only projected-gradient information and can improve over full-dimensional Nestero acceleration in oracle complexity is technically nontrivial.