Reproduce “Sobolev Regularized MMD 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
Chenyang Tian, Bharath K. Sriperumbudur, Arthur Gretton, Zonghao Chen
The authors introduce SrMMD, which moves particles using a witness function regularized by its gradient under the current particle distribution and its RKHS norm. For a two-dimensional Gaussian-mixture benchmark, SrMMD reached final MMD 0.01093 after 200 updates, versus 0.11187 for MMD flow, but each iteration took longer. Its convergence guarantee requires a hard-to-verify source condition and concerns MMD or KSD.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.