Reproduce “Adversarial Likelihood Estimation With One-Way Flows” 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
Omri Ben-Dov, Pravir Singh Gupta, Victoria Abrevaya, Michael J. Black, Partha Ghosh
The authors recast adversarial training as likelihood estimation, using importance sampling and generator entropy to estimate an energy model’s partition function. Their one-way flow evaluates generator densities at generated samples without an invertible network. On CIFAR-10, one-sample FID is 42.4 versus 61 for WGAN-GP; on a synthetic 25-mode grid, the method covers all modes. Image calculations use approximate Jacobians, not exact.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.