Particle Based Variational Inference

Particle-based variational inference (ParVI) is a family of methods using sets of weighted particles to approximate complex probability distributions, primarily aiming to improve the efficiency and scalability of Bayesian inference. Current research focuses on refining algorithms like Stein variational gradient descent (SVGD) through techniques such as dynamic weight adjustments, accelerated gradient flows, and preconditioning, often leveraging optimal transport theory and Wasserstein gradient flows. These advancements enhance the accuracy and speed of ParVI, impacting diverse fields including Bayesian optimization, machine learning, and robotics by enabling efficient inference in high-dimensional and complex models.