Proximal Sampler

Proximal samplers are algorithms designed for efficient sampling from complex probability distributions, particularly those arising in challenging optimization and machine learning problems. Current research focuses on improving the convergence rates of these samplers, exploring variants like stochastic proximal samplers and those incorporating logit mixing or annealing schemes to handle high-dimensional data and non-smooth potentials. These advancements offer significant improvements in computational efficiency for tasks such as direct preference optimization, Monte Carlo integration, and generating node embeddings in graphs, impacting fields ranging from reinforcement learning to robotics.