← back to paper
arxiv: 2605.05179 · 2 revisions
Estimating the expected output of wide random MLPs more efficiently than sampling