Reproduce “Squared Neural Families: A New Class of Tractable Density Models” 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
Russell Tsuchida, Cheng Soon Ong, Dino Sejdinovic
SNEFY constructs densities by squaring neural-network outputs and normalizing them through pairwise activation integrals. Exact normalization is available only for selected activations, representations, and base measures with closed-form integrals. On photometric-redshift estimation, SNEFY n=32 reported held-out negative log-likelihood −2.195, versus −2.156 for conditional flow, with 179,748 versus 1,413 parameters and longer compute time; this comparison is not parameter-matched.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.