Ye He, Krishnakumar Balasubramanian, Sayan Banerjee, Promit Ghosal
The paper provides finite-particle convergence rates for the regularized Stein variational gradient descent (R-SVGD) algorithm, offering non-asymptotic bounds and guidance on parameter tuning.
This research focuses on improving the Stein variational gradient descent (SVGD) algorithm, which is used in machine learning to optimize complex functions. The authors introduce a regularization technique that reduces bias and provides more accurate results when working with a finite number of particles. They establish mathematical bounds that show how quickly the algorithm converges to the desired outcome and offer guidance on how to set various parameters to balance accuracy and computational efficiency.
