Reproduce “Bayesian Compression for Deep Learning” 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

Christos Louizos, Karen Ullrich, Max Welling

Bayesian Compression reduces model storage by pruning groups of units and their connections, then estimating weight precision from posterior uncertainty. On LeNet-5-Caffe for MNIST, BC-GHS reports 419× weight-storage compression at 1.0% error, versus 228× for Sparse VD at the same error; the original reports 0.9%. Pruned models show 2.9× to 3.2× lower GPU energy in batches of 8,192; precision was not used.

Summary by alphaXiv.

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