Reproduce “SoU-RHS: Bayesian Structured Sparsity for Compact Architecture Recovery in CNNs” 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

Rafael Mouallem Rosa, Julyan Arbel

SoU-RHS prunes whole filters from overwide convolutional neural networks using Bayesian horseshoe shrinkage. It sharpens filter-specific shrinkage so weak filters are easier to distinguish from supported ones, then prunes using a posterior-probability rule. Across three plotted datasets, retained-filter counts are more stable across starting widths and thresholds; this supports, but does not prove, recovery of a uniquely optimal architecture.

Summary by alphaXiv.

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