Reproduce “DisARM: An Antithetic Gradient Estimator for Binary Latent Variables” 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

Zhe Dong, Andriy Mnih, George Tucker

DisARM estimates gradients for Bernoulli latent-variable models without continuous relaxations. It pairs antithetic binary samples and analytically averages over the auxiliary randomness used by ARM, yielding an unbiased estimator with variance no greater than ARM’s at the same computational cost. In a FashionMNIST linear VAE, its test 100-sample bound was −252.30 ± 0.05, better than ARM and REINFORCE LOO but below RELAX.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.