New theory quantifies convergence for Langevin-regularized SVGD

This paper introduces a new theoretical framework for understanding Langevin-regularized Stein Variational Gradient Descent (SVGD). The research establishes quantitative convergence guarantees to the target distribution and uniform-in-time propagation of chaos for this method. The findings detail how both the Stein and Langevin components contribute to dissipating relative entropy, leading to exponential convergence rates under certain logarithmic conditions. Additionally, the paper presents finite-time analyses for propagation of chaos, offering explicit bounds in Wasserstein distance and kernel Stein discrepancy. AI

IMPACT Provides theoretical underpinnings for optimization methods used in machine learning, potentially improving convergence and stability of training algorithms.

RANK_REASON This is a theoretical computer science paper published on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• CatalyzeX Code Finder for Papers
• DagsHub
• Fisher information
• Gotit.pub
• Hugging Face
• Langevin
• Langevin-regularized SVGD
• ScienceCast
• Stein
• Stein Variational Gradient Descent
• T_2 inequality
• Wasserstein

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →