Accelerating Low-Rank Matrix Estimation with Preconditioned Gradient Descent

The author presents a method using preconditioned gradient descent to accelerate the convergence of low-rank matrix estimation, ensuring minimax optimality and immunity to ill-conditioning.