2023/03/24 by Minh‐Ngoc Tran, Tran, Minh-Ngoc, Paco Tseng +3 · 1 citation
Mathematics · Computer Science · Decision Sciences · #Statistical Methods and Bayesian Inference #Bayesian Methods and Mixture Models #Probabilistic and Robust Engineering Design
paper · pdf · doi:10.48550/arxiv.2303.13930
The Mean Field Variational Bayes (MFVB) method is one of the most computationally efficient techniques for Bayesian inference. However, its use has been restricted to models with conjugate priors or those that require analytical calculations. This paper proposes a novel particle-based MFVB approach that greatly expands the applicability of the MFVB method. We establish the theoretical basis of the new method by leveraging the connection between Wasserstein gradient flows and Langevin diffusion dynamics, and demonstrate the effectiveness of this approach using Bayesian logistic regression, stochastic volatility, and deep neural networks.