Tag
Proposes a variance-reduced zeroth-order Langevin sampling method for non-log-concave distributions, establishing the first non-asymptotic convergence guarantees, and applies it to inverse problems with score-based generative priors.