Reproduce “General Uncertainty Estimation with Delta Variances” 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
Simon Schmitt, John Shawe-Taylor, Hado van Hasselt
The authors introduce Delta Variances, which estimate how uncertainty in a model’s parameters affects a downstream quantity by combining its parameter gradient with a covariance estimate. In a reduced-resolution, retrained GraphCast evaluation, they found broadly similar proxy quality to alternatives at lower inference compute, while ensembles had the highest average quality; the proxies do not measure ground-truth epistemic variance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.