Constrained and Composite Sampling via Proximal Sampler

Researchers have developed a novel 'proximal sampler' algorithm for constrained and composite sampling problems, which are fundamental to Bayesian inference and machine learning. Unlike existing methods that rely on computationally expensive projections or barrier functions, this approach uses only minimal oracle access to enforce feasibility. The method reduces sampling problems to uniform sampling over lifted convex sets, potentially offering significant efficiency gains for complex probabilistic models without requiring detailed knowledge of constraint geometry.