Reproduce “Regularized Stein Variational Gradient Flow” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ye He, Krishnakumar Balasubramanian, Bharath K. Sriperumbudur, Jianfeng Lu
The authors introduce regularized Stein variational gradient descent (R-SVGD), a deterministic method that corrects SVGD’s kernel-smoothed updates with a regularized inverse. Its mean-field flow approaches Wasserstein gradient flow as regularization parameter ν tends to zero, under a score-regularity condition. In a one-dimensional Gaussian-mixture simulation, smaller ν yields lower log-MSE than SVGD across three functions and both sweeps, but this does not establish general finite-particle superiority.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.