Reproduce “Balanced Training of Energy-Based Models with Adaptive Flow Sampling” 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
Louis Grenioux, Éric Moulines, Marylou Gabrié
The authors introduce flowMC-EBM, which trains a normalizing flow alongside an energy-based model and uses calibrated flow proposals to improve sampling between modes. On a two-mode mixture, its mode-weight estimate stays near the target two-thirds, unlike ULA-EBM. On a 64-dimensional four-Gaussian mixture, its R-hat is 1.05 versus 3.10 for NT-EBM; these synthetic results do not establish performance on natural data.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.